Today's AI is 'more artificial than intelligent', eprivateclient Annual Trust Dinner hears

11/12/2018 Katie Royals,

Artificial intelligence (AI) is a “fashionable” subject everyone talks about, but “very few of us actually know what it means”, Vladimir Dimitroff, Principal at Synpulse UK, told an audience of senior trust professionals at PAM Insight’s annual eprivateclient Trust Dinner held in London on 27 November 2018.

Despite this, AI is nothing new. Humans have been “trying to play god” throughout history, aiming to create something similar to them, “in their own image and likeness”. The approaches have varied across the ages, but the nearest to what we now view as AI emerged shortly before World War 2, Mr Dimitroff explained.

This began with “logical constructs”, using only pencil and paper whereby people would write equations that could solve themselves. This was followed by writing chess moves and emulating game decisions. The “thinking behind it was purely abstract, mathematical” at that stage.

However, shortly after the war, computers began to enter into life, which led to scientists taking this intelligence to a place we now recognise as AI. After some time, the algorithms they were creating began to be sophisticated enough to play the game of checkers and even defeat humans. “Note, that was in the mid-50s, decades before IBM’s Deep Blue defeated Kasparov”, Mr Dimitroff reminded the audience. Such successes, however, this did not evolve rapidly, as the early applications of AI were not immediately “seen or applied, or proven to be beneficial”, Mr Dimitroff explained.

AI moves and develops in “waves”, as interest levels peak and trough. Throughout the years there have been surges of development followed by an “AI winter”, where people lose interest for a while. After a few ‘cold’ years, however, it becomes the “latest new thing” again and developments begin to occur once more, enabled by a newer generation of information technology.

Mr Dimitroff continued to tell the audience how using so-called decision trees and classification trees in ‘expert systems’ for areas like medical diagnostics, which was “back then a revolutionary application”, in turn “propelled the development of machine learning,” now at the “core of AI”. He explained that artificial neural networks (ANNs), a key machine learning construct, began making decisions and became self-propagating. They are now able to back-propagate across the network to improve the final result.

When speaking about the development of these machines, Mr Dimitroff explained that they can be trained in a supervised or an unsupervised way, with the difference being that the former gives preliminary outcomes, while the later does not. Additionally, machines can be trained using “reinforcement learning”, which is similar to “training a puppy.” In reinforcement learning, the algorithm aims to earn ‘rewards’ (score points) by producing the correct results. 

Machines can eventually become almost ‘too clever’ and begin to ‘cheat’ and earn all the points without producing useful results. About such imperfect algorithms Mr Dimitroff joked, “today’s AI is more artificial than intelligent”. This was particularly evident in early versions of virtual assistant bots like Amazon’s Alexa, Microsoft’s Cortana and Apple’s Siri whose abilities made them curiosity gimmicks of little value. With time, however, most of these are now very useable and are widespread in everyday life.

One particularly effective use of AI is image and pattern recognition, as machines can either be trained or self-trained to achieve levels of accuracy that are “often greater than humans”. Mr Dimitroff gave the example of the routine daily procedure of recognising signatures. Machines can do this “more accurately and more predictably” than humans, even when equating for factors such as more stress and using the wrong pen.

Arguably more complex, AI is now used for facial recognition, with nearly all smartphones being able to focus on a face, and some cameras can even focus specifically on eyes. 

He continued to speak about the current uses of AI within the financial services industry, where documents are scanned and sorted on a regular basis. Using optical character recognition (OCR), banks can now scan documents with complex data on them and machines can apply NLP (natural language processing) to classify them with relative ease. In the past there has been an army of professionals trying to read and interpret these texts, but now computers can do this and are able to classify seemingly fuzzy and unstructured data. This is currently one of the “ultimate” applications, “but it is already possible today and will become everyone’s daily work”.

Looking forward, Mr Dimitroff addressed the question of whether AI is going to replace jobs. He said that the situation is often seen as “nirvana or apocalypse,” with everyone fearing the apocalypse. However, he quoted a Harvard University professor who said that “machines will not take the job of a human but a human with a machine will take the job of a human without one”.

Mr Dimitroff concluded that the next phase will be “augmented intelligence,” where AI augments human intelligence and enables individuals to perform previously impossible tasks, working in ‘hybrid’ mode rather than replacing humans. 

Share with Linkedin Share with Twitter

Poor   Average   Good   Excellent