Cognitive Computing: Another look at cognitive tasks
Knowledge formation presents challenges for both humans and computers of course. It is characteristic of knowledge that it can arise out of multiple paths: It can be explicitly trained, captured through deduction from previous knowledge, inductively generated to create net new insights, etc. It is only in recent years that computers in the lab have demonstrated the ability to put forward novel insights about patterns in data sets. And it is important to note that the data sets themselves have been selected by humans. Computers have huge difficulties in projecting patterns and making inferences that human babies do by simply using the innate capabilities of their brains. While computers can certainly ingest massive quantities of digital information—at a rate and a quantity totally beyond human scale, what they can accomplish with that information is limited and still very much dependent on human facilitators.
To talk about memory, long term and working, is to talk to a core strength of computers. There is simply no contest when comparing the fragile human memory facilities with what computers can do through their various storage options. However, the human facilities are inherently multimodal in a way that computers have yet to demonstrate. If you need to put together smells, sounds, colors, spoken language inflections, body language and instant historical time referents for all of these elements, you will need humans in the process.
Our coming intelligent machines are truly challenged where evaluation and judgment are concerned. If by evaluation, you mean “reading” accident report photos in an auto insurance claim context, then maybe computers can be trained to do a first-pass evaluation of the severity of damage. Or in the FinTech context, robo-advisors may be successful in handling simple, pre-defined portfolio investment strategies. But outside tightly constrained industry-specific contexts, open domain evaluation and judgment of complex issues are beyond today’s most advanced computer-based systems.
Reasoning is something that both humans and computers can accomplish, but again everything depends on the nature of the context in which the need for reasoning occurs. If the problem context can be reduced to logical steps, stages, even predictable loops and cycles, the computer can be a speedy, reliable, repeatable resource. However, if the reasoning must have the flexibility to take totally novel events, processes and/or data sources into consideration, the human engine will outperform.
Problem solving and decision-making are cognitive processes that are very close to reasoning in the current state of human/computer capabilities. Highly routinized situations, tightly logical “cook books” for decision guidance—this is a good recipe for computer engagement. Customer calls in, no internet service after power outage, computer voice says re-start your modem. On the other hand, geopolitical strategy or experimental cancer treatments for specific patients are areas where the computer can provide no more than backup data, and humans need to be in control.
Finally, in the area of communicating through language, we are at an interesting tipping point with the growth of computers’ natural language processing, voice and language recognition and natural language generation capabilities. Apple, Google, Microsoft and others have invested billions over decades now to bring us to a world of voice-operated devices and information systems of all kinds. Despite the disappointments and comedy dragging down not-ready-for-prime-time systems in the early releases for smartphones and the more recent disasters associated with enterprise chatbot projects gone wrong, the overall trend is up, and improvements are coming with increasing speed. So our computers are talking to us now and will broaden the conversation year by year. The subtler aspects of human communications, however, will continue to be driven by people. We are built to run language, after all, as computers are built to run numbers.
In looking at the seven cognitive processes as an alternative framework for designing cognitive computing systems, your mileage will vary. But the approach outlined here offers a systematic way to look at cognitive tasks and human/computer partnerships. Leave the hype and technology arguments to the media and the research labs. Look carefully at how people actually work and push hard to understand how much computers can (or can’t) help.