The road ahead for AI: engendering trust
The next stage of innovation in computer science will be not just to make software run faster but also about how to get software to emulate more human-like traits and ultimately become truly trustworthy
The most fundamental progress that technology has made over the past few decades is a dramatic increase in computing power coupled with a reduction in the cost of storage. Such progress has enabled us to run algorithms and code over swathes of data and at a fraction of the cost, thus yielding what some may consider more insights. Consequently, we are able to find patterns in large amounts of data.
This, by some limited definition, is construed as intelligence and has been demonstrated in specific classes of activities such as detecting diabetes from retinal scans, driving cars, playing games, etc. In these cases, software can emulate human behaviour and perhaps even exceed human capabilities. But the fact is that we have not yet invented any new magic wand to make software “think” on its own or have any sense of “consciousness”, in the way humans do.
The limitation is grounded in the fact that software still operates, for the most part, on the paradigm of Gigo (garbage in garbage out), meaning it mostly only does whatever we tell it to and is biased, based on whatever data we train it with. Give it incorrect data and it will pick up incorrect patterns.
But the vision and aspiration of computer science has always been to build a machine that can “accurately” emulate human “intelligence”. However, we still have a long way to go before we can build a software system that is functionally capable the same way a human is. Software “intelligence”, though defined in academic circles, has been fairly misused (and misunderstood) in popular reporting, giving people the impression that software can do just about anything a human being can. This notion is fundamentally misguided. We are not there. Or at least not yet. The gap between software and the human mind is still massive.
Consequently such gaps do not necessarily support claims that artificial intelligence (AI) and automation will replace hordes of jobs in the Indian IT industry and ultimately lead to mass unemployment. This industry employs provides direct employment for a few million people in India, a significant fraction of whom belong to our expanding middle class. Therefore, as various commentators have begun to call out the “threat” of automation to this industry there is considerable (and reasonable) fear whether the rise of machines spells doom and gloom for a section of the Indian middle class.
But I do not believe we are there yet. Or at least not to the extent people fear. Automating away this labour force is not an easy task. Terms like “robotic” automation are not sufficient replacement for the complexity and range of tasks that this labour force performs. It is very tempting to prognosticate a future where silicon (computers) will rule supreme over carbon (humans). However, as a computer scientist, through my work and study over the past few years in the IT/technology industry and about how businesses consume technology, I have gained a genuine appreciation for the unique value and contributions human beings bring to the table in building trust and enabling businesses the world over. Automating these unique attributes with machines will necessarily require emulating the same value humans deliver in businesses.
The role of professionals in the IT industry, the populace whose work is under threat from automation, has not yet been sufficiently understood nor has it been given its due credit. These professionals perform a range of activities, from understanding customers” business requirements to writing code, maintaining software systems and processing transactions. Working in teams, they perform a variety of different actions touching several information systems, coordinate with each other, navigate uncertain or changing scenarios, report to their managers, take bottomline responsibilities for their respective systems, are accountable and serve customers across the globe. All these activities engender a deep sense of trust and confidence among their customers the world over.
But this is also true for each individual. All humans, I argue, even in the most seemingly “robotic” tasks bring to bear abilities that do not make them merely “robotic”. Conceptually every human being has a “brain stack” comprised of layers that include the ability to execute deterministic tasks, intuition, experience, handling (unexpected) variations, common sense, and an ability to identify and fix the unusual. These are fundamental traits that all humans share, though, admittedly, some more than others. No matter the task each human has the potential and ability to apply these traits in any situation without necessarily being to do so a priori. It is a combination of these traits inherent to each one of us that engenders trust and confidence in us among our colleagues.
All businesses run on such trust and confidence in their stakeholders — be they employees or vendors.
Therefore, attempting to replace any human with a “robotic” system is akin to trying to hire a human being who can only ever execute deterministic tasks but lacks intuition, common sense, experience, and an ability to handle unusual situations!
Hence, any attempt by “intelligent” software at automating all that work done by a single individual or an entire team of humans must involve recreating the same trust and confidence. This, in-turn, necessarily requires addressing the whole gamut of these activities that humans perform on a regular basis and with similar human traits, but in software.
Therefore, the next stage of innovation in computer science will be not just to make software run faster but also about how to get software to emulate more human-like traits and ultimately become truly trustworthy (like humans).
Rohan Murty is junior fellow at the Society of Fellows at Harvard University.
This is part of a series of articles in Mint’s 10th anniversary special issue that look at India 10 years from now. The entire list of articles can be found here