Once associated with negative connotations - such as unemployment due to job automation and industry redundancy, or sci-fi movie plot-lines to ‘destroy the world’ - AI is now widely accepted, adopted and better understood by people outside of the technology sector. An array of accessible mainstream AI applications means it has been seamlessly integrated into many elements of our daily lives. Whether through a virtual assistant that helps to book doctor appointments; an email service that can accurately suggest the end to a sentence or predict the next word in text; or a system that suggests content we might like to stream and watch - artificial intelligence is now pervasive and growing in capability.
As a result of mainstream AI applications like these, there can be a false perception that the technology has, for the most part, mastered human language. In truth, AI has proven to be successful in natural language-based use cases, where it is designed and trained for a specific purpose. This is because the problem definition is clear and narrowly defined in scope, which enables the use of scripted techniques, and or statistical neural network-based approaches. But there has been less success with general conversation capability, or with techniques to understand the meaning that humans express through their use of natural language. We are far from AI-based systems that can reason the way humans do.
The current nascent state of the technology is evident even in the leading consumer products. Alexa Siri, and other voice and digital assistants use AI to interact with users. These products, though widely adopted, can execute only a limited number of tasks and are incapable of offering meaningful user experiences based around conversation. Instead, they enable users to execute tasks and actions through them but not to engage in dialogue with them or grow their understanding over time. The limitations of approaches used by these products are glaringly obvious. Delivering the value and experience consumers want from natural language-based experiences, requires us as practitioners and industry leaders to make new breakthroughs. It seems that cognitive development may reveal answers.
Last year, Josh Tenenbaum, who leads the Computational Cognitive Science Lab at MIT, announced plans to study the minds of children to inspire ‘the next big innovation in AI’. Tenenbaum claims that by studying how young children see the world and learn, we can apply similar techniques to develop an AI that will be truly ‘intelligent’ - or, at least, intelligent in a way that is more useful to consumers.
In a talk at the EmTech MIT conference last September, Tenenbaum discussed the idea in depth. "Why do we have all these AI technologies, but fundamentally no real AI?" Tenenbaum said. "We have machines that do useful things we used to think only humans could do, but none of these systems are truly intelligent, none of them have the flexible, common sense [of] . . . even a one-year-old."
Artificial intelligence that acquires and processes information in the same way as children in their earliest stages of cognitive development has the potential to change our lives and our interaction with machines existentially. An AI that has been trained to learn directly from a specific user; emulating their thought processes and model of the world to become, essentially, a digital sounding board; would be life-changing and invaluable. To do this well requires greater emphasis on small data and individualised human models with a greater need for dataset design to get suitable training sets for machine learning models. In my opinion, only once AI becomes human-centred and multi-disciplinary in its definition, design, research and development, will humanity be able to truly reap the full benefits it can offer.
It’s unsurprising that few noteworthy breakthroughs like this have been made, because the growth in natural language, as an interface, is dominated by Apple, Amazon, Microsoft and Google. Each has a legacy way in which it generates revenue and uses its voice-based experience to reinforce this. The rapid growth in smart home devices is a land grab to own the user; and how they access digital products and services. Data shows that many of these devices end up merely as voice-activated speakers and kitchen timers and offer little more than convenience. It doesn’t seem that true product-market fit has been achieved by any. It’s a form of corporate protectionism through subsidy, more than it is innovation.
Modern-day innovators, like Tenenbaum and numerous others, perceive a significant responsibility to build an artificial intelligence that is truly human-centric to its core; in both what it does and how it learns. I believe that it’s not only our duty, but is also the largest opportunity to create and capture value; crucially, it’s the path that will force us all to be more creative and make technological breakthroughs. By asking better questions - such as “how can we use natural language to add meaningful value in people’s lives?”, instead of “how do we sell more ads and capture more data?” - you will get better answers.
Artificial intelligence has the potential to extend human life and evolve our own intelligence. Rather than purely being a functional layer within our lives, it should be helping us see our blind spots; enabling us to become more capable over time and more able to figure out better solutions to the challenges we face in life. A technology that could objectively make links between the different elements of our lives - human relationships, financial issues, health problems etc. - through learning from each individual user, would help humanity all over the world, regardless of the distinctly varied issues we all face day-to-day. Based on the progress I’m seeing with cognitive science-based approaches, this future may not be so far away.
Tom Strange,CEO and Founder, Constellation
Image Credit: Enzozo / Shutterstock