Artificial Intelligence: Where do you go to, my lovely?

Waiting for the Robots

Everyone has wondered about the robots. When will they be here? The first science fiction book I read was Logan’s Run. Written in 1967, the novel portrays a dystopian future where, in order to balance the limited resources available with the needs of the population, everybody has to voluntarily die when they reach the age of 21. To make that pact agreeable to all the citizens, their focus, until age 21, is on the pursuit of pleasure. Nobody who is going to die at age 21 has the time nor wants to be saddled with the responsibility of bringing up kids, so that job, along with many others, is given to the robots. Coincidentally, also in 1967, the first automatic teller machine was installed outside a Barclays Bank in North London. Now, this was a real robot! Well, kind of. It offered 24-hour convenience to customers. It didn’t call in sick, nor come to work with attitude, and it didn’t even think of demanding a pension. If you were a bank teller, it seemed pretty clear that it was angling to take your job. Soon, ATMs were everywhere.⁽¹⁾ Imaginations went wild. In science fiction, the robots became smarter, they learned to think, then they took over the world. Fast forward 52 years… surprise – there are now more bank teller jobs in the U.S. than when the ATM was introduced. Of course, it turned out that tellers could be re-tooled for higher and better uses, such as foisting loans and credit cards onto customers, perhaps purely to help them in the pursuit of pleasure?

Where are they Today?

So, where are the robots? Arguably, they are here. They just haven’t taken your job… yet. By way of explanation, this takes us to another anniversary. In the summer of 1969, fully 50 years ago, the lunar module of Apollo 11 landed on the moon in the Sea of Tranquility. Critical to this stunning achievement, and making calculations way faster than any human, was the on-board Apollo Guidance Computer (AGC). You likely know the gist of how this story evolved. By 1998, a simple hand-held student calculator was roughly as powerful as the AGC. Today, a late-model smartphone has over 100,000 times the processing power of the AGC, one million times more memory, and seven million times more storage.⁽²⁾ And it’s not just brute computing power. Your iPhone has (some) artificial intelligence built in. It can predict the next word you are typing (sometimes) and does a very good job of speech recognition. This is not just Siri listening and then telling you the weather. Many people aren’t aware that you can dictate an email and your iPhone will do the speech-to-text task for you.⁽³⁾ In 2016, Microsoft’s AI speech-to-text machine achieved a milestone, whereby it became as accurate as a professional human transcriptionist (and it surpassed human accuracy the following year).⁽⁴⁾ You likely also know that a computer can play a Masters-level game of chess. OK, you say, impressive, but it has all been carefully programmed by humans… where is the intelligence? How about getting a computer to play computer games when it hasn’t even been told what the rules are? Google’s Deep Mind AI team has done precisely this. They equipped the AI with two things: the first is that the computer can remember and learn (from the prior rounds of the game that it has played), and the second is an objective to maximize the score. Equipped with these powers, the computer figures out the rules and improves its strategy over time. After trying 49 classic Atari games (remember Space Invaders?), the computer achieved a skill level of at least 75% of that of a professional gamer on 29 of the games. Interestingly, in a couple of cases, the AI developed winning strategies that human players had never thought of. AI-based improvement by having the machine learn from past cases, then by making predictions itself with human-directed “coaching” is key to progress. A professor at Georgia Tech has developed an AI teaching assistant, which, along with 8 real teaching assistants, answers student queries online. The AI “bot” was built using IBM’s Watson platform and was “fed” 40,000 questions and answers about the course that had been asked by real students and answered by teaching assistants over the prior few years. Initially, human assistants reviewed the AI bots’ responses and improved them, where necessary, before releasing the answers. Within a couple of months, the AI bot was able to post accurate enough answers completely on its own.

This process is not without its problems. Microsoft developed a chatbot called Tay that it released on Twitter. Within 24 hours and some 90,000 tweets later, by learning from other participants, Tay had started to flirt online and developed racist and sexist traits, prompting Microsoft to end the experiment. Efforts are underway to have AI bots read books and watch movies with good moral messages, so as to develop human-like constraints and abide by social boundaries. Good luck with that!

Where Do You Go to, My Lovely?

… where do you go to, my lovely
When you’re alone in your bed?
Tell me the thoughts that surround you
I want to look inside your head
– Lyrics by Peter Sarstedt, 1969

Where may AI be headed? With more power, more speed, more memory, and more learning every year, AI can only get better. AI is already pretty good at identifying cancer in a mammogram. With every newly-confirmed cancerous mammogram, the computer learns and gets better at analyzing the next mammogram. This computer learning concept is being broadly applied for autonomous driving, evaluating credit card applications (yikes – the loan officers, not the tellers, may be out of a job this time!) and much more. You may have come across AI-generated art, such as poetry, music, drawings or 3-D sculptures.⁽⁵⁾ Much of it is pretty clunky, to be sure, but to be able to stand in front of a mirror and see an image, not of yourself, but rather a rendering of you in the way that Vincent Van Gogh may have drawn it, …then Picasso, …then Salvador Dali is pretty neat.

One of the key things that Google (for evaluating websites and YouTube) and Facebook (for social media posts) are feverishly working on is using AI to discern the nature, accuracy, and bias, of websites, blogs, online articles and posts – with a public service objective. Google has to remove inappropriate videos from YouTube, and Facebook needs to identify hate speech and fake politically-motivated accounts. This brings us back to the song lyrics above (they serve a purpose, after all!). Peter Sarstedt had a love-interest in mind when he penned his lyrics in 1969 (the same year as the Apollo lunar landing). Today, we may be on the cusp of a new era – a computer, by analysing what you read or write, can look inside your head. “Fed” enough of what you read, write, and view, it can coldly assess your innermost views and your biases – soon, perhaps, with better accuracy than your spouse. Peter Sarstedt, who died in 2017 (coincidentally the 50th anniversary of the first ATM), probably can’t fully rest in peace. Given all he wrote, a computer can now look inside his head. Now, stop me before I get going on George Orwell’s Nineteen Eighty Four… a dystopian future – here today? FWG