Artificial intelligence and siren songs
Montse Hidalgo, a journalist with Spanish daily El País, published an article (link and pdf in Spanish) that reflects many of the concerns I expressed at a recent OECD forum — and those of much of the machine learning community — about the hype, sensationalism, misinformation and alarmism of much news coverage on the subject.
I have noticed that many people confuse machine learning with artificial intelligence: machine learning is already here, and it works: a company I advise, BigML has some 50,000 customers in 120 countries that use it to process more than 16 million complex tasks. Companies all over the world pay to use tools that dramatically reduce entry barriers to the automation of sophisticated tasks in clearly defined scenarios. Machine learning is not artificial intelligence, as Tom Dietterich, one of the founding fathers of machine learning, co-founder and chief scientist of BigML, pointed out a few days ago in El País:
“Machine learning works well for stable problems, when the world is highly predictable and it is easy to collect large amounts of training data. In problems where every situation is unique, that technology is unlikely to succeed. “
Is it simply that artificial intelligence sounds sexier and less geeky than machine learning? The problem is that we make the leap to talking about machines capable of thinking like humans or better than them, of replacing government with algorithms, murderous robots, etc, etc. This is bullshit. And there is even a course on the subject.
The real problem with this artificial intelligence hype is that unreasonable expectations often lead to erratic adoption processes, as reflected in the well-known hype cycle outlined by Gartner. Machine learning went through a long winter when computers were not powerful enough to implement the models being developed, and could happen again if, once Moore’s law has put things in their place, we insanely inflate expectations. Talking about whether robots are going to take over humanity is like worrying about overpopulation on Mars.
Machine learning is undoubtedly something we need to understand, the new “proficient in Excel” on our CVs. Machine learning should already be taught at all levels, as a part of the environment in which we live, because we live surrounded by machines, and these are capable of learning. Will the development of machine learning lead to the loss of jobs? Undoubtedly, the jobs eliminated will be those dedicated to stable tasks, in predictable environments, with very specific rules and restrictions and capable of generating large amounts of data for analysis. They will replace jobs that, by their nature, should not be done by humans, and undoubtedly, will do them better, more predictably, more quickly and with fewer mistakes. It is not a substitution, it is a liberation, although obviously, some prefer not to be liberated if the alternative is not being able to earn a living. If you want jobs in the future, invest now in machine learning, robotics, and most of all, in education. The digital transformation of education is what keeps me stuck to my job after no less than 27 years: I definitely don’t want to miss that process!
But for the moment, let’s focus on leveraging the technology that is already here and that provides the dramatic improvements that come when a machine learns from data analysis and performs a task. Let’s forget the hype that says, “buy this technology and that’s it”: just collecting the data now stored in relational models, transforming and preparing them so they can be used in machine learning models is a long, painful and complex process that will consume 80% of the effort of any project of this type.
That does not mean that we should not do it, on the contrary, we have to start as soon as possible, because although long and rocky, the road will lead to a pay off, and if a competitor gets them before us, it could push us out of the market, making us irrelevant, obsolete, non-competitive. Machine learning is not about murderous robots, nor intelligent machines, nor philosophical dilemmas: just mathematics and coding.
It would be nice to talk about this without the bullshit and the philosophy and based on more rational expectations. While some spend their time talking about science-fiction, others are making real, tangible progress, and learning to be more efficient in today’s world.
(En español, aquí)
