IEEE Future Directions
Related Articles
Artificial Intelligence has progressed enormously, partly through the creation of new algorithms, mostly by leveraging on processing power and endless availability of data. Insights coming from the study of brains (plural intended, since we are learning from brains as simple a worms and as complex as ours), neuronal circuitry, are going to push even further the progress in the next decade.
I feel we have crossed a first thresholds, the one of usefulness: we now have a variety of concrete applications of AI that are pushing the industry to invest and deliver further products in an ever self-fuelling spiral.
At the same time we are still deeply rooted in weak artificial intelligence, very good in very specific areas. We are still far from AGI, Artificial General Intelligence. Re-tuning AI that is working really well in a specific area to a different one is a costly endeavour.
In the “Forrester AI Readiness Report” out of 717 business surveyed 40% are planning to use AI solutions to improve their business, however the industrialisation of AI is proving to be a bottleneck.
This is one of the reason to work on ways to create adaptive models, evolutionary algorithms that are able to reconfigure themselves to address different areas.
We already have machines (software) that is able to learn through looking at its context and by creating a context into which it can explore evolution path. We are just starting in this area: as I mentioned economic pressures are pushing in this direction. An Artificial Intelligence that can create Artificial Intelligence.
Would this be the way to finally create AGI? May be not, but clearly this is a way to broaden significantly the field of applications.
Notice that evolutionary selection forces may be applied on these self generated replicants, where evolution is both in the direction of addressing broader contexts and in dealing in a different way. This is what has been happening in Nature over billions of years of evolution and like in Nature we can expect an acceleration in features creation, as more and more complex components become available.
Nature took 4 billion years to deliver what we see (and what we are) today. Computers are much faster. Additionally the selection process can be made much more efficient. By setting the stepping blocks for AI self evolution we can expect a boundless, and fast, evolution.
This is also one of the reasons for concern, as voiced by several scientists, including Hawking.
However the issue is much bigger, as it is well described in the clip here: the problem, if a problem exist, does not lie with Artificial Intelligence, rather with evolution as consequence of replication and selection. “Stupid” algorithm, under selective pressure may highjack more intelligent algorithm as result of the selective pressure. Bacteria cannot be considered significantly smart, yet a bacterial species resulting from evolution is “smart”, actually one can claim it is way smarter than the complex organisms it may infect!
Something to think about…
