The spread of inaccurate terms, many times by business activities, gradually make developers, manufactures, sellers, users and apprentices oblivious of the true meanings behind some new concepts. Cognitive technologies, cognitive applications, cognitive services, cognitive systems, are new terms popping in recently across the media. It seems that the cognitive is becoming popular as marketing label for some high-tech products and services, and maybe it will spread quite in the same way that neural and intelligent mushroomed in the technological cosmos. What the pioneers and the notorious companies come to say about these new concepts certainly will influence subsequent views, newcomers and those that popularize technological achievements in the news.

It is not commonplace to find gross mistakes, and the situations mostly convey misused terms, non-applicable concepts or forced extensions. Terms like neural and intelligent are sometimes borrowed to give new colors to old items, without careful considerations, naively adapted and superficially presented. This may confuse the non-specialists, which often contribute for the consolidation of the wrong view. The misuse or abuse of these concepts has been questioned by specialists. It’s opportune to make a similar review of mentions of cognitive items.

Here are samples of what some companies are saying about the cognitive:

– “Cognitive technologies are applications and machines that perform tasks that previously required human intelligence.” , from How cognitive technologies are transforming our capital markets, by Darshan Shandarana, Fuad Faridi and Christina Schulz. McKinsey Insights, July 2017.

– “Technologies able to perform tasks traditionally assumed to require human intelligence, are known as cognitive technologies.”, from Cognitive technologies: The real opportunities for business, by David Schatsky, Craig Muraskin, And Ragu Gurumurthy. Deloitte Review 16, 2015.

– “When we refer to cognitive technology, cognitive automation, or artificial intelligence, we are really talking about an algorithm, or chains of algorithms, that enable software to absorb information, reason, and think in ways similar to human beings.”, from Harnessing the power of cognitive technology to transform the audit, by M. Macaulay, R. O’Donell, V. Swaminathan. KPMG Portfolio, 2017

– “Cognitive computing refers to systems that learn at scale, reason with purpose and interact with humans naturally. Rather than being explicitly programmed, they learn and reason from their interactions with us and from their experiences with their environment.”, from Computing, Cognition and the future of knowing, by John E. Kelly III. IBM Research & Solutions: Computing Cognition White Paper, 2015

The McKinsey and Deloitte examples portray the cognitive as ‘the way that humans do things’, while for KPMG and IBM what counts is that humans learn and reason, and to be cognitive would be doing the same in a similar (although not necessarily identical) way.

By reading the reports and portfolios above cited, one realizes that they agree that cognitive technologies employ machine learning  and logical reasoning, plus something else that performs “in ways similar to human beings”. However they don’t point which ways are these, neither how they are similar to the human ways. This lack of details, one could guess, is maybe due to the difficulty in identifying and summarizing the common aspects of these “human ways”  in a single view valid for every application, as it would be desirable when introducing the idea. Therefore, because of this dependence with the application considered, it is natural to expect that the  “human ways” should be discussed case by case, and this is exactly what we see in the documents above: many examples of how to realize the cognitive in several applications, but without  elucidation of what makes the application a cognitive one.

Nevertheless, we can’t simply blame the authors for not going deeper in explaining what is meant by “cognitive”.  In fact, there’s a lot of disagreement even among scientists and specialists about what is cognition. And the lacking of a clear notion about cognition generally doesn’t constitute a major problem for the class of services and products that many of these companies offer. By now, products and services that really require a precise comprehension of what is cognition are still in the realm of research, testing, conception or even fiction. The current market is yet rehearsing the next steps, which will bring self-driving vehicles, autonomous robots and drones, fully engaging conversational interfaces, smart cities/ houses/ facilities/ and appliances, for instance.


By looking into the  set of envisioned applications and trends potentially involving the cognitive, one could identify two groups comprised by the so-called cognitive applications: the cognitive agents and the cognitive tools.  The cognitive agents are those that  embed cognition to improve their autonomy, while cognitive tools are those designed to enhance or augment the cognition of its users.

Things like self-driving vehicles, autonomous robots and virtual agents capable of high-level interaction with humans, are potential candidates to be considered as cognitive agents. The requisites are to embed cognition and to be capable of reaching operational autonomy. Cognitive tools, on the other hand, could be things like advanced interactive diagnostic tools, interactive recommendation systems, and real-time language translating tools with interactive feedback. The requisite is to provide an interface between the user and the application domain that extends the user’s cognitive capacity and improves the user’s autonomy in dealing with the targeted issues.

Notice that in both cases the presence of cognition is a requirement, to be fulfilled by the agent or by the user, respectively. Therefore, in order to understand the cognitive, cognition is a concept that must  be somehow clarified.

As mentioned previously, this is a controverted concept and although we expect to   consider it carefully in future posts at this blog, we will now provide a brief and simplified definition of cognition just for the sake of clarity and to have some immediate reference:

Cognition refers to a system of processes that yield the build of knowledge and shape its use.

Looks a remarkably simple definition in view of the aforementioned controversy, but this is illusory. In fact, it is rooted in other likewise tricky questions:

  • What is knowledge;
  • What does it mean to build knowledge, and how this takes place;
  • How the use of knowledge is shaped.

Although these questions demand careful considerations which are topics left for further posts, here we rather proceed using the definition without saying more about them. They don’t affect substantially the remarks we will make about to the way that companies are employing the cognitive.

The problems coming from the misuse of the term cognitive are generally related to the  supposed role of cognition in the application. If a company says that it sells a cognitive product or service, it is revealing to identify where cognition is present in order to find if it refers to a cognitive tool or a cognitive agent.

If it is a cognitive tool, it provides knowledge to the user, which is in this case the cognitive part. Applications pertaining to this class usually performs analytic tasks, with the addition of adaptivity, learning, and reasoning. Cognitive tools are not autonomous, they rely in the user to give rise to effects on its operational environment.

The cognitive agent, on the other hand, must work autonomously, and the role of cognition  here is to enhance its autonomous performance,  giving the agent an improved capacity of making successful decisions, of being more robust and build upon its past experience. The agent is now the cognitive part, and cognition is a process subsumed in its functioning.

Notice that in no instance we have mentioned that the cognitive application should do something in the way that humans do, either for a cognitive agent or a cognitive tool. In fact, the crucial point with cognition is not reaching human performance, but improving autonomy in some sense, for the agent or for the user, respectively. And this autonomy has to do with the quality of the prediction of situations and the correction of errors. The search for “performing in ways similar to human” is a typical target of artificial intelligence, which paved a different path in the conceptual and methodological landscape. Intelligence has to do with task performance optimality, while for cognition just task accomplishment as desired with increased autonomy of the agent (or the user). Following an anecdote, if you make a robot watchdog, you don’t require it to perform like a watchman.   We finish by saying that

to be cognitive does not imply on being intelligent,

although this doesn’t mean that making cognitive applications is easier than making intelligent ones.

What does it serve AI small