If you’re a regular blog reader, you’re probably already aware that when it comes to artificial intelligence, its current state of development is severely misunderstood. Sadly, however, this matter is not helped at all by the many cash-eying marketers who are still promoting the myth that you can purchase sentient tech products.
Take Oral B, for instance. The entire campaign for their Genius X toothbrush is based around the notion that it offers a full slate of quasi-magical AI features. Yet peel away the hype from what is actually objective fact and what users will be left with is an overpriced feedback system, talking a big game but with nothing to back up its claims.
Yes, true artificial intelligence, as it turns out, remains a developer’s pipe dream, as researchers are nowhere near full comprehension of the human mind.
But what then is the actual state of AI technology today — which really should be referred to as machine learning, by the way — and what should companies expect from the most probable future ahead? In this article, Software Planet Group will take a candid look.
It Started with a Game
To understand where we are today, it is helpful to have a look at a leading AI enterprise: the originally British, now Google-owned project DeepMind.
The theory behind the project was certainly an interesting one, that because human beings excel at logic and maths, then perhaps if a computer could be taught how to do the same, all of the rest would then simply fall into place.
DeepMind’s original creator, Researcher Demis Hassabis, was heavily inspired by IBM’s Deep Blue computer. After all, in 1997, the computer had become the de facto world chess champion, having beaten the human title holder in a staggering six-game rematch.
Though in 2016, DeepMind’s AlphaGo program took this one step even further, by achieving the same with Go — the most complex game known to mankind — had the board been changed a fraction to a simpler, smaller variety, DeepMind would have suffered a humiliating defeat.
This serves to illustrate an important defining point: that computers are still limited to the data we feed into them.
Advances in Language Processing
More recently, therefore, bearing in mind these limitations, AI researchers have largely shifted towards more practical applications, like so-called NLP or natural language processing tools.
Think of Apple’s Siri or the Amazon Alexa. Despite their occasional mishaps, a mere decade ago, such assistants were reserved to the realm of science fiction. Fast-forward a couple of years, however, and we are making steady progress in interpreting more complex sentences.
Consider as an example the following linguistic construction: “in an overwrought, self-obsessed and oh-so annoyingly preachy way, the book — against all odds — somehow manages to work.”
Previously, virtually every algorithm in existence would have wasted no time concluding that this was a negative review. Yet by contrast, today’s NLP systems are programmed in such a way that they pay special focus to the pivotal concluding portion — correctly marking the critique above as indeed, a favourable one.
Advances in Computer Vision
Another area which in recent years has been showing remarkable improvement is image classification, or “computer vision,” as it is known as today. Like the name implies, this refers to a computer’s ability to correctly identify objects as well as colours and people, and their positions in relation to others.
As one might expect, this was extremely difficult to achieve, yet by attempting to closely imitate the complexity of the visual cortex, researchers can make AI systems make sense of a cluster of pixels.
Better yet, by pairing computer vision with modern-day NLP tools, computers can now behold an image and express in words what they are “seeing.”
This is set to make an enormous impact especially in the field of healthcare. In oncology, for instance, we are already able to provide patients with fully automated blood cell counts, help doctors to spot infections and reduce medical expenses.
As the technologies mentioned above will continue to develop further, you should not be surprised at all if your translator starts offering “live listening.” Equipped with such a feature, for example, by simply holding up your phone to someone speaking in a foreign language, you could then hear it played back to you in your own particular tongue.
Similarly, machine learning is expected to completely revolutionise radiology. This application in particular, however, will require vast amounts of training data, as there are thousands of things to look for and accuracy is more crucial than ever. For this reason, automating the entire process is still a ways away.
Only As Smart As You Make It
The natural conclusion, therefore, is that at least for now — and probably throughout our lifetimes — so-called artificial intelligence will remain as “smart” as we make it.
Though there is certainly no denying that a degree of perceived intelligence exists, this is more like an amateur illusionist who tries desperately to convince you of his magic, as it is certainly not organic and does not function like the human mind.
Nonetheless, the future does look undeniably promising, so if natural language processing, computer vision or intelligent feedback sound in any way appealing to your company, please get in touch and SPG will be happy to help!