In the late 80’s and early 90’s I was pursuing a post-graduate degree in Artificial Intelligence. It was a bit of a renaissance for the field as some significant breakthroughs occurred in learning algorithms. The availability of cheaper computing also fuelled a range of new applications. So what has changed in the past 25 years?
Certainly not much compared with the transformation to our lives that portable internet connected devices has made. Certainly not much when compared with the breakthroughs in DNA sequencing and their transformations to medical science.
The progress is like that with Nuclear Fusion. Plenty of promise but nothing that has dramatically changed our daily lives. And even more like Nuclear Fusion, the actual understanding of the science has not really progressed much. There is still simply no clue as to what “consciousness” is and how it manifests.
Yes the engineering has progressed and there are some cool applications like Google Translate and other voice recognition. However, these are largely limited to very constrained application domains. The systems themselves have very little ability to actually understand what is being said. Let alone being able to meaningfully participate in a conversation.
Copying the Brain
A Special Report in the IEEE Spectrum magazine (June 2017 edition) sets the scene for a “moonshot” type series of projects to “copy the brain”. One article “An Engineers’s Guide to the Brain” (pp24 -25) is virtually indistinguishable from diagrams in my under-graduate and post-graduate theses of some 25 years ago. We have learned very little about how information is represented in the brain over that time. We have learned much less about consciousness itself.
This is not to say that Artificial and Machine Intelligence engineering has not progressed substantially. There are many very bright pathways in the future for this field however it is know amongst seasoned practitioners and researchers that every generation there are predictions of Artificially Intelligence rivalling Human Intelligence in the next 20 years! These prognostications have occurred regularly since 1920.
Good at Playing Games
The feats of Artificial Intelligences in beating humans at the games of Chess and Go not withstanding we are so far away from this goal that it is highly likely that a child born today will not see this outcome in their life time. And that is likely a life extraordinarily extended to at least 100 years. See also When NASA got AI.
Killer Drones
Of course military applications for drones making a battlefield decision of life or death have been in the media, see Killer Drones. This is a terrifying prospect not so much because the machines are making the decisions but because this type of research will have very little “peer review”. By necessity this type of military research is kept secret so we as a scientific community will be unable to determine the validity of the science and engineering being used. Whether or not this is Artificial Intelligence or just some algorithms being called such will not be public knowledge.
Chat Bots and Credit Scores
It would also appear that the main function of machine learning in the Financial Services industry are for automated credit scoring and loan eligibility evaluation and for “Chat Bots”. These Chat Bots are able to handle the first few queries of a customer’s chat before they are inevitably transferred to a human operator to deal with the vagaries of real world problems faced by a bank’s customers. Both these applications are great starts in the use of these Artificial Intelligence techniques but hardly industry transforming at this stage. See also Jobs vs Robots.
IMHO we can look forward to steady and incremental improvement in technology through the adoption of Machine Learning and Artificial Intelligence techniques but it will be a long time before the AIs take over.