What AI can and can't deliver
The dangers of 'deep stupidity'

17th November 2017

There is no doubt that AI is smarter than it has ever been before and promises to disrupt many walks of life. At the same time, debate continues to wage over whether it can really challenge human intelligence or whether it despite having intelligence in many areas, suffers from 'deep stupidity' in others?

We look at some of the limitations of machines including understanding feeling and insight, inspired by Professor Mark Bishop's talk at the recent 'Minds Mastering Machines' conference, in London, where he argued that only by appreciating the limitations of AI can we exploit its full potential.

The need for human-provided context
Fully comprehending human language involves understanding the context in which the language is used, which holds multiple challenges for AI machines.

In many languages, certain words have multiple meanings or are described using replacement vocabulary which relies on wider general knowledge, such as 'Googling' something, meaning 'searching the internet' or 'requesting an Uber' meaning 'calling a taxi'. Such technicalities create ambiguity for machines aiming to respond to humans without context of the conversation.

In order to understand human language comprehensively, AI would rely on a bank of global general knowledge that can be retrieved to supplement the understanding and context of data.

Emotional intelligence
While AI-enabled machines are able to communicate with other machines and in many instances can liaise with humans, they still cannot mirror humans’ sensitive and appropriate approach to situations. In the area of customer service, for example, technology can provide information in response to questions, but does not necessarily possess the communication and empathy skills, which are necessary to develop relationships and make customers feel valued.

AI can provide information to solve complex problems, but only when they have rules or systems to adhere to, unlike the human brain which has the flexibility to approach each situation uniquely and apply problem-solving skills based on intuition.

Bishop demonstrated this view with the example of a book that reached the price of $23,698,655.93 on Amazon when set by a machine without common sense to realise that this was an unnaturally high figure. Similarly, in the healthcare field, doctors are frequently presented with complex and rare cases, which demand the application of vast knowledge and use of common sense, which may differ in regards to each patient and defy set rules or ideas.

Whilst AI is making strides towards human parity in certain activities, there are other areas in which it is lagging behind and a human approach is still superior. In order to apply AI effectively, we need to appreciate the nature of the task in hand and choose the tasks it is best suited to. Otherwise, we will be setting it up to fail.