Is AI’s Promise Achievable? – Scotiabank Perspectives
Everywhere you look right now there’s one buzzword: AI. Movies have been made about it. TV cop shows use it to solve crimes. Artificial Intelligence is the latest tool that is supposed to cure society’s ills and give us more free time, but is that promise achievable today?
What is AI?
Machine Learning is the current application of the concept of AI. Machine Learning is computers programmed to “learn” from data sets and associated human preferences in order to anticipate human needs and deliver correct solutions.
Artificial intelligence can be seen in many things you use every day, including SIRI or Cortana, suggestion engines on shopping sites, and even digital radio stations. The number of places you’ll see AI is increasing.
What Can AI Do For Us?
You may remember Watson from Jeopardy! a few years ago. Watson was a computer programmed to learn trivia and the mechanics of playing Jeopardy! in order to see if it could play alongside human competitors. What you might not know, is since then, Watson has been attending Medical School, learning how to diagnose diseases. How good is Watson? Since 2012 when he entered Medical School, he has beaten doctors at a Leukemia diagnosis.
Other notable uses of AI include DoNotPay, an app that helps you beat parking tickets (it has already saved people $9.3 million disputing 375,000 tickets using bots.) Branching out from parking tickets, the company is looking at other simple legal processes it can automate for customers
Nest (and Canadian competitor Ecobee) are learning thermostats that learn from your family’s usage patterns of heat, air-conditioning, and other automated services and deliver your optimal experience to you when you want it. Now imagine a box that anticipates all your other needs, including knowing that your milk is about to go bad and your dog’s prescriptions need to be filled. That’s the promise of AI.
But, Is AI Really That Powerful?
Experts claim that AI is throttled by a few factors: Computing power (being able to come to decisions quickly is of paramount importance. Some hardware isn’t up to the task,) Data integrity (how good and clean the data is. Imported data can have keying errors, missing data, repeated inputs, etc.)and Behavioural. Humans sometimes lie, make decisions against their own interests, or don’t understand choices fully. AI is still worse than a toddler at understanding emotions.
AI progress has not been as fast as other computing breakthroughs and some suggest that it might be slowing down slightly. It’s not purely dependent upon faster chips and extremely clean data. It’s also dependent upon our understanding of how humans’ brains work to solve problems. There’s no computational equivalent to “instinct” yet.
AI has another hurdle which rarely makes headlines: it’s still very dependent upon human interaction and oversight. AI bots left to their own devices without human intervention have had disastrous results, like the Microsoft Twitter bot that became a Neo-Nazi in 24 hours after launch. As much as humans might feel like they can’t predict human behaviour, they are light years ahead of computers.
AI is still wildly expensive for all but the biggest players, so for smaller companies looking to build the next big thing, the cost of operation (hardware, programmers, who are in scarce supply and charge accordingly) remains too high for them.
While AI has eliminated some pains of our modern world so far, AI isn’t ready to fundamentally reshape our world as we know it, yet, but it is making bigger strides every day toward a more Jetson-like future.