Author: Ronald Schmelzer

Ron is principal analyst, managing partner, and founder of the Artificial Intelligence-focused analyst and advisory firm Cognilytica, and is also the host of the AI Today podcast, SXSW Innovation Awards Judge, founder and operator of TechBreakfast demo format events, and an expert in AI, Machine Learning, Enterprise Architecture, venture capital, startup and entrepreneurial ecosystems, and more. Prior to founding Cognilytica, Ron founded and ran ZapThink, an industry analyst firm focused on Service-Oriented Architecture (SOA), Cloud Computing, Web Services, XML, & Enterprise Architecture, which was acquired by Dovel Technologies in August 2011.

Do Companies Need a Chief AI Officer (CAIO)?

The short answer to the above question is “no”. But let’s get into the specifics. Clearly for many companies, Artificial Intelligence (AI) and the range of Cognitive Technologies are strategic to their businesses and organizations. Indeed, for many companies, AI is as fundamentally important to their long-term well being as their IT organizations and finances. So, if you have a Chief Information Officer (CIO) in charge of all the information and IT-operation activities in the organization and a Chief Financial Officer (CFO) in charge of all the finances for the business, why shouldn’t you have a Chief AI Officer (CAIO) in charge of all the AI-related activities?

Do you have a Chief Mobile Officer or Chief Internet Officer?

From helping to identify and defray fraud to improving business processes to new products and services that have not previously been possible, AI is enabling a wide class of solutions. However, just because something is important, even fundamentally and significantly so, doesn’t mean that the corporation’s management team needs a shakeup with a new person at the top rung in the organizational structure. Every few decades, a new major technology movement comes that shakes up the economy and shuffles the way businesses work. In the recent past, the most significant of these movements have been the emergence of the Internet and mobile devices as separate major technology trends. There’s no doubt that the Internet pretty much changed everything in most industries from retail to finance to real estate, and similarly there’s no doubt that the emergence of mobile computing was another tectonic shift for those industries and others that might have weathered the Internet revolution.

Yet, did we see an explosion of Chief Internet Officer or Chief Mobile Officer titles? Perhaps at the early stages, companies flirted with those titles as a way to signal to the outside world that it took the Internet and mobile seriously, but now I think you’d be challenged to find any Fortune 1000 company with those titles as part of the top-level “C stack” in the organization. Why is that? Simply put, while there’s no doubting the significance of the Internet, mobile, and other movements such as Big Data and cloud computing to the enterprise, the strategic portion of the organization didn’t need some new high-level executive to add to management team. If anything, companies saw these new movements reinforcing the value of their existing team and adding to the strategic importance of management-level titles such as the Chief Information Officer (CIO) and Chief Technology Officer (CTO).

If AI and Cognitive Technology are Important to You, then Get Your Existing C-Level Team to be “AI-First”.

Much of the reason why companies are creating CAIOs is because they want to signal to the outside world that they find AI so core to their business and mission-critical that they will create and elevate a position within the company that’s solely focused on AI and directly reporting to the CEO. However, it seems both short-sighted and organizationally confusing to use a C-level title as a way of indicating to the outside world the importance of a company’s AI efforts.

Rather than create a whole new C-level executive, companies that are itching to name a CAIO should instead follow the examples of companies that have declared their intent to be “AI First”.  Specifically, over a year ago, Microsoft decided to ditch their “mobile first” mantra, which was questionable in its ability to have the strategic results the company desired, to become “AI first”.  In the speech announcing the move, Microsoft CEO Satya Nadela stated, that Microsoft would create “best-in-class platforms and productivity services for an intelligent cloud and an intelligent edge infused with AI”. This is just as much about putting a stake in the ground for the market and competition to know as it is as a rallying cry for their internal employees and stakeholders. Microsoft even when a step further, creating a dedicated site with a mission-statement, objectives, and ethical considerations in its move to be “AI First”. If anything, this says more than simply appointing someone to an ill-defined CAIO role.

Likewise, in 2017, Google also declared their intentions to be “AI-first”, moving away from a mobile-first and mobile-centric world. Specifically, Google CEO Sundar Pichai stated that Google was moving away from “searching and organizing the world’s information to AI and machine learning.” That’s a very fundamental statement for a company that built its reputation on wrangling the world’s information.  Both Google and Microsoft took action to followup on their words and made major executive changes, but did not appoint a CAIO. Google named a new head of AI initiatives and Microsoft split its development and engineering divisions into a Cloud group, an AI group, and an experiences and devices group. These were all big management changes, and yet no CAIO.

What Would the CAIO Be Responsible For Anyways?

Another thing to consider with regards to the CAIO role: just who and what would the CAIO be managing as part of the role’s organizational efforts? If a CAIO is a chief of something, it needs to be a chief of managing either people, resources, or both. It doesn’t make sense for the CAIO to manage the development team implementing Machine Learning (ML) models for enterprise end user customers because much of where AI and ML implementation happens is within the line of business.  And even if AI development is not happening within the line of business at enterprise end users, then it’s happening within the IT organization and managed by the CIO.  Does it make sense for the CIO to handle all information except AI and ML-related data and information?  Not really. Does it make sense for the line of business managers to handle their line of business except when it has to do with AI and ML? Not really.  So without people or information to manage within an enterprise, the CAIO is useless.

Within a vendor’s organizational structure, the CAIO title makes little sense as well. The CTO usually manages the company’s long-term technology vision and sets strategy for the company as it pursues new business opportunities. It doesn’t make sense for the CTO to manage all strategic initiatives except those relating to AI and ML.  Likewise, the actual product development usually rests with the VP of Product Management or the SVP of some product or development organization. Why would that SVP work on those products that are not AI related, if AI is so critical to the company?  In these cases, if a technology vendor names a CAIO and it also has a CTO, CIO, and VP Product Management or SVP, lots of conflict over product and strategy ownership will occur.  If the CAIO is not managing anyone or crafting a strategy, then what value is that role?

Forget the Title, Go For the Value

Indeed, the best way for a company to signal to internal employees and stakeholders as well as outside customers, investors, partners, and competitors that it is serious about AI and ML is to do the precise opposite of naming a CAIO: not naming a CAIO. Instead, companies that aim to make AI and ML a core part of the organization’s mission, mantra, and message should task each of the existing C-level executives to make AI and ML part of their organization’s mission. The CEO should be “AI-first” and speak in terms of how AI and ML will strategically impact the business. The CIO, CTO, CFO, CMO, Chief of Human Resources, and other strategic executives need to also craft their messages from the lens of AI and ML, if they wish to be portrayed as being strategically focused on AI.  To further this statement, if you see any company naming a CAIO while also retaining the CIO, CTO, CMO, and other relevant positions, you should come to the conclusion that AI is not strategic for their firm or they are confused about AI really fits within their organization.  After all, why would they separate the CAIO responsibilities from the rest of the organization while keeping those other roles intact?  This is a signal that AI is somehow separate from the rest of the business and not core, despite what that firm might otherwise indicate.

Are We Heading to Another AI Winter?

Amongst all this hype and bandwagon jumping on Artificial Intelligence (AI), Machine Learning (ML), and Cognitive Technologies is also a sense of unease. How is it that a technology that has roots going back as far as the beginnings of computing is suddenly now the hot “must have” technology that’s powering ever-more dramatic amounts of money being pumped into a few skyrocketing startups?  The industry has gone through two major waves of AI development and promotion with their own periods of sky-high hype only to sink dramatically back to earth once people realized the limitations of what surely was being hyped as being on the cusp of sentience.  And so here we are again, in the “summer” of this wave’s AI adoption wondering if this will all last, or if billion-dollar unicorns are being funded in an environment that’s sure to pull back the reins of overinflated expectations.

Revisiting the Causes of the AI Winters

An AI Winter is a period of declined interest, funding, research, and support for artificial intelligence and related areas — in essence, a “chill” on the growth of the industry. There have been two major AI winters, each following a period of heated interest, funding, and research growth for the industry. The first wave of AI interest in the 1950s-1970s was followed by the AI Winter in the mid to late 1970s, and the second wave of AI interest in the late 1980s-mid 1990s was followed by a subsequent winter.

In our analysis, the reasons for the AI winters are many: overpromising and underdelivering on AI capabilities (hype beating reality), lack of diversity in funding sources, overcomplicating technologies, not providing enough of an advantage over non-intelligent “status quo” options, and robbing the research pipeline by diverting researchers into industry jobs. Certainly many others have written about AI winters, their causes, and how to avoid them. But increasingly among the dim of interest and overheated expectations on AI we’re starting to hear those in the industry wonder if all this interest has already peaked.

In late May 2018, Filip Piekniewski published a blog post titled “AI Winter Is Well On Its Way” that garnered significant attention in the industry.  In his post, he bemoans the overhyping of the industry and puts facts to the claim that deep learning is not achieving the much-vaunted goals of its promoters.  He states that autonomous driving is starting to hit the real limits of learning and autonomy, and particularly debunks claims that AI will displace knowledge workers in fields such as radiology. In essence, Piekniewski has thrown a big bucket of water on the raging fires of AI hype and promise.  However, is this bold claim true?  Is another AI Winter really well on its way, or are researchers just tiring of industry promotion?

Are Enterprises committed to AI?

The Wikipedia entry on the AI Winters has an interesting take on the phenomena, seeing it from the lens of AI research. The perspective is that the AI winters first starts among researchers, and then spreads to the press, and finally to investors and industry. If all AI Winters follow this pattern, then surely we have something to worry about, as notable AI researchers such as Rodney Brooks are starting to get grumpy. For sure, research happening at universities, research labs, and institutions are important to the development of AI, since we’re still trying to understand and grapple with the most basic understanding of what intelligence really means and how we can make machines more intelligent. But does the buck really stop with AI research, and is AI research the canary in the coal mine warning us of industry pull back?

From our perspective, the buck starts and stops with enterprise adoption. This is not to say that the enterprise matters to everyone — rather, this is solely our perspective as an analyst firm focused on the enterprise. Companies with more than just a modest number of employees and sales are complex machines, having to coordinate the multiple needs of customers, employees, product development, service delivery, investors, partners, shareholders, and others. While research is important to enterprises in that it helps develop competitive advantage with products that are able to continuously meet customer demands, enterprises aren’t committed to research for research sake. Rather, for most enterprises, the question is, “does this technology solve a problem? Do my customers care?”

From this perspective, the question is not is the next AI Winter here, but whether we have even reached the summer yet.  AI is not a discrete single technology, but rather a collection of related cognitive technologies that each address a different aspect of how previously only human cognition or capability could be applied to a specific problem. In the past, only humans could recognize objects that fit into patterns, but now it’s possible to train machines to be very effective at image and object recognition.  To many, image recognition is a “solved” problem in AI, and the applications in enterprise are immediate.  No one can convince companies to stop using image recognition applications because their value has been proven.

Likewise, cognitive technologies are being leveraged to be able to process and generate natural language, handle a wide range of pattern-matching and decision-making tasks, and interact with the environment using sensory capabilities that previously were too complicated to do with traditional approaches. The appetite for investing in these technologies is only beginning for most enterprises, and both internal corporate budgets for machine learning-enabled solutions as well as venture capital money seems to continue to flow to projects that are meeting real business-world problems, rather than pure research.

Saying AI but Meaning Something Else

Perhaps the issue that concerns AI researchers is that the term AI is being used too broadly. AI purists would tell you that the pursuit of anything but “strong” Artificial General Intelligence (AGI) is short-sighted. If you truly want to have the breakthroughs in the ability to create really intelligent sentient machines, you need to solve the hard problems of AI, and not use AI-like “parlor tricks” that are more about big data management and improvements in statistical processing algorithms that leverage powerful computing resources than they are about grappling with fundamentals of what intelligence really means. While that mindset might be correct from the AI researcher’s perspective, it doesn’t make those parlor tricks any less useful to the enterprise.

In a future article, we’ll dive into whether the term Artificial Intelligence might really be the best term to use for enterprises if a general cooling towards adoption of AI technologies starts to come.  However, for this article we’d like you to consider what you really want from an intelligent machine.  At a recent vendor event that Cognilytica attended, the CEO keynote speaker claimed that within just 7 years, we’ll be walking by humanoid robots and won’t be able to tell if they are machines or not. Putting aside whether or not this bit of technology hyperbole was real or not, the reaction of the mostly enterprise-focused audience was telling: laughter.

To many non-researcher lay people, the idea of Artificial General Intelligence (AGI) is rapidly becoming crackpot territory. The shenanigans of Sophia, the fear-mongering of Elon Musk and others, and statements by vendor CEOs who think they are impressing their audiences are making non-technologists wonder if the pursuit of AGI is only for crazy people.  The more that these figures as well as a slew of novelists, bloggers, Hollywood, and podcasters keep pushing ridiculous claims about how soon we are to world domination by sentient robots that will be our overlords, the more we’ll just hasten the pullback from AI in general. The risk to AI is not that AI will underdeliver on industry promises, but rather that we’re being told (or sold) one thing: AI, but being delivered something else. In one breath, we’re being sold a vision of humanoid robots, and in the next breath, we’re told about process automation. The cognitive dissonance is remarkable.

What We Need to Progress AI Research and Keep Funding Going without Risking another AI Winter

So where are we heading? Are we really heading to an AI Winter? The answer is… it depends. Many of you familiar with analysts will know that this is the stereotypical analyst response. Of course it depends, but what does it depend on?

First and foremost, we need to separate the goals of AGI research and continued AI research from the goals of applying AI and cognitive technology needs of enterprises and consumers. Companies don’t need humanoid robots to be able to successfully implement chatbots for customer self-service.  Autonomous vehicle manufacturers don’t need superintelligence to be able to design vehicles that can successfully navigate chaotic streets and avoid accidents. Organizations don’t need sentient systems to be able to build autonomous systems that can handle constantly evolving business processes.

Perhaps we can look at research around AI and the output and outcomes of AI research much like how we approached the space race. The goal of the space race was to put someone on the moon and launch missions to the outer planets and beyond.  Many people did claim we’d be living on the moon by 2001 or colonizing other planets, and those visions helped to power development of enormously valuable technologies.  Those technologies are what’s actually changing our lives today, from Kevlar and Velcro to baby formula and aerogel. Yet it didn’t take living in a space station to get there.  Space research has not stopped and neither has adoption of space research-derived technology.  Similarly, if we can keep our minds inspired by the vision of what AI can become, but our feet planted firmly on the ground for what AI technology can deliver, we can simultaneously keep money and interest flowing to AI research while applying shorter-term AI technologies to immediate needs. In this way, we can avoid the next AI Winter, or delay it for years to come.

Make Way for Cobots

Czech writer Karel Čapek first coined the term ‘Robot’ in his 1920 play R.U.R. about a dystopian future in which artificial, manufactured “roboti” act human-like and perform tasks in servitude to their human masters, only to later to form a rebellion that leads to the extension of the human race. Many of the visions and dreams about what intelligent machines can be and the fears associated with them are in this way related to the origin of the term Robot in this way. Originally envisioned as physical, hardware things, the term robot is used in a wide array of manners to deal with any sort of software or hardware-based automation, whether intelligent or not.

Yet, physical robots are still highly desired in many industries, especially to perform tasks often referred to as the four “D’s”: Dirty, Dangerous, Dear ( or Expensive), and Dull (or Demeaning). These robots operate every day in manufacturing, warehouse, health care, and other situations to perform the tasks that would otherwise be performed by humans with not always positive outcomes. However, to make industrial robots work in a reliable way without causing physical harm to humans, they often must be separated from physical human contact. Operating in entirely human-free zones or within cages that prevent accidental human contact. Or, if they are roaming about in the free world, they are constrained in their strength and capability so that they can’t inflict harm. But constraining physical robots in this way limits their application and power. Companies looking to increasingly automate and enable greater portions of their business that require physical human labor currently need ways to increase the interaction of robots and people without endangering their welfare.

Cobots: Augmented Intelligence Approaches to Physical Bots

Often the industrial robots that might come to mind when you think of physical bots operating in 4-Ds type environments need separation from humans to operate safely. However “collaborative robots”, known by the shorthand “cobots” are meant to operate in conjunction with, and in close proximity, to humans to perform their tasks. Indeed, unlike their more isolated counterparts, cobots are intentionally built to physically interact with humans in a shared workspace. Cobots aren’t actually new. The first examples emerged in the mid 1990s from university research projects and the General Motors (GM) Robotics center in which humans would provide the power to make the machines move while the cobots would provide the control and steering to place objects with precision. In this way, humans were safe because they controlled the power of the robot while gaining all the advantages in assistive capabilities that the machine would provide.

In many ways, cobots are the hardware version of augmented intelligence that we talk about in the software world. Instead of replacing humans with autonomous counterparts, cobots augment and enhance human capabilities with super strength, precision, and data capabilities so that they can do more and provide more value to the organization. Pioneered by robotics company KUKA and further developed by Universal Robots and then Rethink Robotics, cobots have increased their footprint in industrial settings both large and small.

Cobots are also trained differently than traditional industrial robots. Rather than programmed to a specific set of steps using programming tools, many cobots are trained by demonstration. Humans control the bot by physically moving it around, with the cobot remembering the steps and perhaps even the end goal of what is being accomplished, and then repeats those steps, optimizing them to achieve increasingly better outcomes.

Broadening the Application of the Cobots

There are a few really interesting implications of the cobot term and concept. From a technical perspective, it’s interesting that even in heavy industrial applications where traditional non-collaborative robots have been used for decades, cobots are seeing increasing adoption. It’s clear that there are a range of automation and intelligence tasks that previously couldn’t be addressed by dangerous to interact with robots, that can now be intelligently automated with collaborative robotic solutions.  If hardcore robotic adopters like automotive and industrial solutions can find uses for cobots, surely the softer manufacturing, logistics, supply chain, warehouse, and perhaps even retail and consumer industries can find a range of cobot use cases. Will the Starbucks barista of the future be a cobot? From what we can tell, this is an inevitability.

More importantly, the entire concept of the cobot is its cooperation and ability to work in close proximity to humans. These cobots are aimed at augmenting what humans do, to make humans more effective, efficient, and enhanced. Indeed, if the old robotics focused on the 4 Ds’, then the new cobotics focuses on the Three E’s: efficiency, effectiveness, and enhancement.

Since cobots are aimed at helping humans do their jobs better and not replacing them, it’s no surprise that human acceptance and enthusiasm in working with cobots is high. Companies see much greater adoption with cobots working alongside humans than they do with robots that are meant to replace their activities. Many view the term “robot” as scary (much thanks to Čapek for kicking off our fears), but the term “cobot” is more comforting and friendly.

If this is the case, then why are we not using the term cobot to apply to collaborative bot technology that’s software based? We’ve had much concerns about the term Robotic Process Automation, which seems to combine the idea of unintelligent automation with human-replacement robotic concepts. It’s no wonder enterprises get resistance from their workers trying to implement software bots that aim to replace human labor. Instead, as we discussed in our Intelligent Process Automation (IPA) report, we should be talking about cognitive process cobots. Rather than dealing with unintelligent, human-antagonizing robots, we have helpful, human-assisting cobots. Rather than being hardware based, these cobots are software-based intelligent assistants that aim to help humans do their jobs better. They provide the three E’s described above: efficiency, effectiveness, and enhancement. What’s not to love about your friendly process cobot?

Changing an Industry by Changing Terminology

Sometimes all that’s necessary to achieve new outcomes and benefits it to change the terminology that’s used. Terminology impacts many aspects of technology adoption: how people perceive your place in the market, who your competition is, how the technology can be applied, what concerns and issues may arise, and even things like price point and implementation complexity. If you call a car a person transporter instead, you might have a different opinion about what the car is used for, what alternatives to the car there are, and perhaps how many people can fit into it. If you call a bus a multi-person vehicle, similarly your impressions of what the bus is changes.

Similarly, using the term “robot” might not be a good fit for all autonomous or intelligent applications. Simply calling something a robot conveys something in the mind of the user, adopter, and buyer. Calling it something else, even if the technology stays the same, might bring different concepts to mind. Since the term robot is overloaded with many connotations both positive and negative, we should shift how we refer to intelligent assistants and augmented intelligence solutions. Those solutions that are meant to replace humans with intelligent assistants, and perhaps even be isolated from humans to operate with greatest safety can be called robots, if that term applies. And conversely, those applications where intelligent assistants are meant to be used in an augmented fashion, working side-by-side with humans to assist them with tasks can be called cobots, if that term fits. In this way we can separate the fears and desires from using one word with that of another and focus on the application and not the specific terminology.

Smart Speakers? Meh. Intelligent Assistants? Yeah!

Ever since the launch of Amazon’s Echo device in 2014, its seems that every month brings a new development in dedicated devices that process voice commands and perform actions. However, what exactly are these devices? The popular media calls them “smart speakers” or “voice assistants” or “intelligent personal assistants”, but these words describe very different concepts. A smart speaker conjures up a primarily output oriented device that aims to replace keyboard or button interaction with voice commands. Yet, that seems to be a particularly trivial application for the significant investments and competitive posture that Amazon, Google, Microsoft, Apple, Alibaba, Tencent, Baidu, and countless others are taking. After all, why are all these vendors so aggressively marketing and promoting these devices if all they do is allow you to play Taylor Swift on vocal demand or let you ask about the weather?

Clearly there’s a bigger play here than simply the smart speaker. The smart speaker is just a way to initially get their product into a larger number of households and businesses and get people comfortable with using these devices. The real play is something bigger than just a speaker you can control with your voice. The power is not in the speaker, but in the cloud-based technology that powers the device. These devices are really low-cost input and output hardware that are a gateway to the much more powerful infrastructure that sits at the major tech companies’ data centers. The device itself is the giveaway to this. You can even build your own full-featured conversational device for just a few dollars.  So let’s dispense with the clearly ill-fitting term “smart speaker”. It belies the real power of these devices.

Not Smart Speakers. Intelligent Conversational Assistants.

If you ask Amazon, Google, Microsoft, Apple, and others, you know that playing music, games, and responding to simply queries is not the end state of their vision for what these conversational gateway devices will be. This week’s demonstration of Google Duplex at Google I/O 2018 clearly shows the power of what an intelligent conversational assistant can truly be. Rather than just being passive devices, intelligent conversational assistants can proactively act on your behalf, performing tasks that require interaction with other humans, and perhaps soon, other conversational assistants on the other end. The power is not in the speaker device.

Indeed, where exactly is the device? The device (speaker) is completely missing in the Google Duplex scenario. We don’t see a device because a device is not necessary here as the devices are just gateways to the real activity that’s happening in the data centers.  The conversational agent is acting completely behind the scenes from Google’s data center interacting through voice-over IP (VoIP) telephone lines with a human on the other end.

So, why are devices needed at all if they’re just gateways? They’re needed because they provide the user interface to the cloud-based intelligence services. Without a device, the only way to access these services is through a web, desktop, or mobile interface. But this is inefficient. Amazon wasn’t truly the first to bring voice-based assistants to market. Apple had them beat by over three years with Siri, and Google introduced their voice-based assistant in Android just a short while after. What made Amazon stand out though with their Echo devices is that the mobile phone was eliminated entirely.  Rather than activating the device through a phone, you can simply speak in the comfort of whatever activity you’re doing and trigger intelligent capabilities. Basically, the value of the device is in its hands-free mode of interaction, but the intelligence of the device is in the back-end infrastructure.

How Intelligent Are These Devices?

Earlier this year, Cognilytica announced the creation of our Voice Assistant Benchmark. The purpose of the benchmark isn’t to test the natural language processing (NLP) or natural language generation (NLG) capabilities of the devices. Nor is the intent of the benchmark to see what sort of skills these devices can perform. We know that better NLP/NLG means the ability to handle a wider range of voices, accents, languages, and speaker characteristics, and more skills mean more single-task capabilities. Those are all “table stakes” as far as we’re concerned.

If the power of the devices is not in the device itself, but in the back-end intelligence that gives these devices real capabilities, then we need to test to see how intelligent that back end really is. Can the conversational agents understand when you’re comparing two things together? Do they understand implicit unspoken things that require common sense or cultural knowledge? For example, a conversational agent scheduling a hair appointment should know that you shouldn’t schedule a hair cut a few days after your last hair cut, or schedule a root canal dentist appointment right before a dinner party.  These are things that humans can do because we have knowledge and intelligence and common sense. Yet as it stands and as we demonstrated in our initial benchmark, neither the Google Home nor Amazon Echo nor Apple Siri devices can answer the question “what’s larger, the sun or the earth?” Are these devices you’d trust running your life? Not yet.  But, we aim to help move things in that direction.

The Implications of an Intelligent Conversational Assistant

In the not-so-distant future, intelligent assistants will be everywhere. We’ll be interacting with them daily in both our personal and business lives. We’ll be chatting with assistants in our homes, and also interacting with other people’s and business’s conversational agents. In a future where everyone will have a personal electronic virtual assistant, we’ll have them do everything from messaging friends when you’re putting together a birthday party, to scheduling all the logistics for that party, to dealing with inbound calls from late attendees who can’t make it. Soon enough, as dependent as we are now on our GPS systems from keeping us from getting lost and our mobile phones for keeping us always connected, we’ll be dependent on these intelligent assistants for keeping our lives in order. This is just an inevitable direction of where things are heading.

However, there’s a downside to the use of intelligent assistants. In a recent article in Verge, experts bemoan the fact that humans will want to know if they’re talking to a robot or not. Clearly people will be frustrated by the earlier generations of intelligent assistants as they make frustrating mistakes. Yet, there’s an even darker potential outcome. Criminals and mischief makers can use voice assistants to tie up phone lines, cause retail “denial of service” attacks by scheduling fake appointments, cause harm by faking information to people to get them to leave their houses or otherwise tie up resources. In the future, we’ll need a sure-fire way to make sure that we know who the speaker on the phone is, what their intentions are, and how real the requests are. The future, which is really here now, is that we can’t believe anything we see or hear. This makes verifying reality incredibly important in an AI-Enabled Future where intelligent assistants are part of our everyday lives.

Chasing the Elusive Machine Learning Platform

If you have been following the breathless hype of AI and ML over these past few years, you might have noticed the increasing pace at which vendors are scrambling to roll out “platforms” that service the data science and ML communities. The “Data Science Platform” and “Machine Learning Platform” are at the front lines of the battle for the mind share and wallets of data scientists, ML project managers, and others that manage AI projects and initiatives. But what exactly are these platforms and why is there such an intense market share grab going on?

The core of this insight is the realization that ML and data science projects are nothing like typical application or hardware development projects. Whereas in the past hardware and software development aimed to produce functionality that individuals or businesses could individually run or control, data science and ML projects are really about managing data, continuously evolving learning gleaned from data, and the evolution of data models based on constant iteration. Typical development processes and platforms simply don’t work from a data-centric perspective.

It should be no surprise then that technology vendors of all sizes are focused on developing platforms that data scientists and ML project managers will depend on to develop, run, operate, and manage their ongoing data models for the enterprise. The thought from these vendors is that the ML platform of the future is like the operating system or cloud environment or mobile development platform of the past and present. If you can dominate market share for data science / ML platforms, you will reap rewards for decades to come. As a result, everyone with a dog in this fight is fighting to own a piece of this market.

However, what does a Machine Learning platform look like? How is it the same or different than a Data Science platform? What are the core requirements for ML Platforms, and how do they differ from more general data science platforms? Who are the users of these platforms, and what do they really want? Let’s dive deeper.

What is the Data Science Platform?

In our earlier newsletter piece on Data Scientists vs. Data Engineers, we talked a bit about what data scientists do and what they want to accomplish with the technology to support their missions. In summary, data scientists are tasked with wrangling useful information from a sea of data and translating business and operational informational needs into the language of data and math. Data scientists need to be masters of statistics, probability, mathematics, and algorithms that help to glean useful insights from huge piles of information. A data scientist is a scientist who creates hypothesis, runs tests and analysis of the data, and then translates their results for someone else in the organization to easily view and understand. So it follows that a pure data science platform would meet the needs of helping craft data models, determining the best fit of information to a hypothesis, testing that hypothesis, facilitating collaboration amongst teams of data scientists, and helping to manage and evolve the data model as information continues to change.

Furthermore, data scientists don’t focus their work in code-centric Integrated Development Environments (IDEs), but rather in notebooks. First popularized by academically-oriented math-centric platforms like Mathematica and Matlab, but now prominent in the Python, R, and SAS communities, notebooks are used to document data research and simplify reproducibility of results by allowing the notebook to run on different source data. The best notebooks are shared, collaborative environments where groups of data scientists can work together and iterate models over constantly evolving data sets. While notebooks don’t make great environments for developing code, they make great environments to collaborate, explore, and visualize data. Indeed, the best notebooks are used by data scientists to quickly explore large data sets, assuming sufficient access to clean data.

Indeed, data scientists can’t perform their jobs effectively without access to large volumes of clean data. Extracting, cleaning, and moving data is not really the role of a data scientist, but rather that of a data engineer. Data engineers are challenged with the task of taking data from a wide range of systems in structured and unstructured formats, and data which are usually not “clean”, with missing fields, mismatched data types, and other data-related issues. In this way, the role of a data engineer is an engineer who designs, builds and arranges data. Good data science platforms also enable data scientists to easily leverage compute power as their needs grow. Instead of copying data sets to a local computer to work on them, platforms allow data scientists to easily access compute power and data sets with minimal hassle. A pure data science platform is challenged with the needs to provide these data engineering capabilities as well. As such, a practical data science platform will have elements of pure data science capabilities and necessary data engineering functionality.

What is the Machine Learning Platform?

We just spent several paragraphs talking about data science platforms and not even once mentioned AI or ML. How are data science platforms relevant to ML? Well, simply put, Machine Learning is the application of specific algorithms, additional unsupervised or supervised training approaches, and learning-focused iteration to the large sets of data that would otherwise be operated on by data scientists. The tools that data scientists use on a daily basis have significant overlap with the tools used by ML-focused scientists and engineers. However, these tools aren’t the same, because the needs of ML scientists and engineers are not the same as more general data scientists and engineers.

Rather than just focusing on notebooks and the ecosystem to manage and collaboratively work with others on those notebooks, folks tasked with managing ML projects need access to the range of ML-specific algorithms, libraries, and infrastructure to train those algorithms over large and evolving datasets. ML Platforms help ML data scientists and engineers discover which machine learning approaches work best, how to tune hyperparameters, deploy compute-intensive ML training across on-premise or cloud-based CPU, GPU, and/or TPU clusters, and provide an ecosystem for managing and monitoring both unsupervised as well as supervised modes of training.

Clearly a collaborative, interactive, visual system for developing and managing ML models in a data science platform is necessary, but it’s not sufficient for an ML platform. As hinted above, one of the more challenging parts of making ML systems work is the setting and tuning of hyperparameters. The whole concept of a machine learning model is that it’s a mathematical formula that requires various parameters to be learned from the data. Basically, what machine learning is actually learning are the parameters of the formula, and basically fitting new data to that learned model. Hyperparameters are configurable data values that are set prior to training an ML model that can’t be learned from data. These hyperparameters indicate various factors such as complexity, speed of learning, and more. Different ML algorithms require different hyperparameters, and some don’t need any at all. ML platforms help with the discovery, setting, and management of hyperparameters, among other things including algorithm selection and comparison that non-ML specific data science platforms don’t provide.

What do ML Project Managers Really Want?

At the end of the day, ML project managers simply want tools to make their jobs more efficient and effective. While we have written earlier that not all ML is AI, and perhaps some of the ML approaches are used primarily for non-AI predictive analytics, those seeking to add true intelligence as part of their mission need the same capabilities regardless of how ML is being applied. The real winners in the ML platform race will be the ones that simplify ML model creation, training, and iteration. They will make it quick and easy for companies to move from dumb unintelligent systems to ones that leverage the power of ML to solve problems that previously could not be addressed by machines. This is the ultimate vision of ML as applied to AI: make systems autonomous, intelligent, and generate knowledge and action that otherwise would require human capabilities.

ML platforms that enable this capability are winners. Data science platforms that don’t enable ML capabilities will be relegate to non-ML data science tasks. Vendor who pretend that their business intelligence, data analytics, big data engineering, programming-centric, or other tools are rebranded AI / ML platforms are in for a rude awakening. We know who you are, and no, you are not an AI / ML platform vendor. Stay tuned for our big report on Data Science and Machine Learning Platforms as we sort out who is doing what in the ML platform space, which data science platform vendors are the ones worth paying attention to in the ML space, what is necessary functionality for ML platforms and what is not, and who is starting to win the race for marketshare in this constantly evolving, but significantly attractive market.