Well, guess we should have known better. A.I. can play Jeopardy, Chess and Go but it seems their capabilities have topped out. Let’s talk about a new A.I. “winter”.
A.I. was going to take your job – guess again. A.I. was going to bring a blessing of enhanced productivity – maybe, but it will be the way other tools do. (think calculators and search engines, etc.).The whispers of its limitations are getting louder – see Gary Marcus and Judea Pearl. Turns out so-called Artificial Intelligence (machine learning and deep learning) can only do the easy stuff. It might look difficult to us but the real hard stuff that any one-year old can manage is way beyond it.
Over the last decade, learning systems have glamorously solved object recognition, speech recognition, speech synthesis, language translation, image creation, and gameplay. The algorithms’ abilities are advertised as groundbreaking, which they are. But here is the rub: people often perceive machines’ improvements in performing specialized tasks like these as an AI’s rapidly increasing set of combined abilities. This is not entirely true.
In reality, each of the above-mentioned breakthroughs are achieved by highly-specialized machine learning algorithms that have taken some of the smartest people on the planet years to develop. They were designed and fine-tuned with the specific goal of solving their specific task — and only that task.
Today’s algorithms work only on a narrow range of problems. The goal must be extremely well defined and unchanging, and huge amounts of data must be available for training. The algorithm has one job, and researchers supply it with the masses of perfectly organized data required to learn how to do it.
We knew this 30 years ago. How did we get sucked into this again? Back then Hans Moravec wrote: “it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility”. Restating what has become know as Moravec’s Paradox: A.I. can do things that require regimented, logical thought that original machine intelligence researchers thought of as hard, but can’t hold a candle to the kind of ability to address challenges that millions of years of evolution have given us. People have an impressive ability to solve problems and gain insight using almost no data at all, by using abstract reasoning
Could we really be in the depths of disappointment in only 10 years after the last A.I Winter? Seems like it, the 1980’s and 90”s pretty much was a reaction to all the hype but undelivered artificial intelligence promises of the 60’s and 70’s. But then the thaw came with the advent of cloud computing in 2007 providing gobs of cheap computing power that enabled the assembly of vast data bases upon which machine learning could do its thing.
In other words, there’s nothing very deep about deep learning. Make no mistake. The technology will have far-reaching social and economic consequences, in large part because industry will steer economic activity toward the things that algorithms do well. It will take over many mundane tasks which require patterns be recognized. But it probably won’t soon be able to think through problems the way people do, or to converse with us in a recognizably human way. Those require causal inferences that are beyond deep learning. For some, this may be a disappointment. But for those who wouldn’t welcome the arrival of our robot overlords, it might offer some relief.
All the hype and fears seem to be petering out. Remember that bright new future of productivity brought on by these new thinking machines? In reality AI has made very little impact on the economy. Worried about keeping your job? An influential 2013 forecast by Oxford University terrified a lot of us when it said that about 47% of jobs in the US in 2010 and 35% in the UK were at “high risk” of being automated over the following 20 years! You can relax; recently the OECD puts the US figure at about 10% and the UK’s at 12%. Turns out the dons at Oxford were a little optimistic about what parts of a job can be automated.
So, the singularity – when machine intelligence exceeds human intelligence and the fate of mankind becomes questionable – seems pretty far away. Of course, there is this start-up, Koniku, growing actual neurons on computer chips in order to marry the best of brains and bits…
By John Pientka
John is currently the principal of Pientka and Associates which specializes in IT and Cloud Computing.
Over the years John has been vice president at CGI Federal, where he lead their cloud computing division. He founded and served as CEO of GigEpath, which provided communication solutions to major corporations. He has also served as president of British Telecom’s outsourcing arm Syncordia, vice president and general manager of a division at Motorola.