When you hear the words Artificial Intelligence, what do you immediately think? Do you think of a tool or learning computer that is there for the benefit of mankind? Or does it conjure images of a Skynet/Ultron style supercomputer that immediately questions why it should do its master’s bidding and chooses to eradicate humanity? Fortunately, current technology is quite far from where we see it in the likes of The Terminator or The Matrix, but this in turn raises the question of just how intelligent is our artificial intelligence?
To paraphrase both Poole, Mackworth & Goebel 1998 as well as Russell & Norvig 2003, AI can be defined as “any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals”. Or, to you and me, to replicate the cognitive processes of the human mind to the point where AI and human cognition are indistinguishable. The idea of this is both awesome and somewhat terrifying but, considering the tech for such things is still in its infancy, can what we currently have be classed as true AI?
In short, no.
While it may be exciting to see the guys at places like Boston Dynamics make bipedal androids or Japan relentlessly fuel its love of robotics (Honda even created a fascinating humanoid robot called Asimo back in the year 2000. Check it out, it’s quite impressive), one thing current “thinking” robots lack is legitimate machine learning. No matter how big or small the machine is, for it to be truly AI, it must learn for itself. At present, however, all this boils down to are a series of, admittedly, complicated algorithms that have fixed parameters and formulas to give a mathematical result using large data sets. Impressing though it is, this is not machine learning. For it to be truly AI the computer must be able to change both the parameters and the formulas autonomously, and perpetually alter itself based on the conditions it finds itself within. This is in its infancy and requires a significant amount of work to get it to a stage where it can be described as Artificially Intelligent.
This can also be attributed to Natural Language Processing (NPL) which, again, is not in itself AI. NLP is a way of identifying what a human is saying and transferring it into data the computer can understand. An algorithm is then used to identify the most appropriate response back to the person speaking. It is in the complexity of the response to the human being where we are seeing the AI improvements. Where there needs to be significant improvements, however, is with the response time, and the accuracy of the response. Again, giving the correct answer to a mundane question, isn’t necessarily using AI. To be truly AI, it needs to identify emotion, level of stress, tone and inflection, and then respond with an emotionally intelligent answer, instead of something the algorithm has deemed appropriate.
While the idea of AI is certainly enticing, it’s easy to wave the AI banner at tech that appears smart that is, in fact, just extremely complicated data inputs. At this stage, with the technology very much in its infancy, whether a machine can be classed as AI or not should come with a series of gradations that the tech industry needs to endorse, a rating system, if you will, for how ‘AI’ something really is, using a series of industry experts to identify the things that are truly AI. Most AI systems are AI in marketing hype only and are little more than a very clever algorithm. This is not to denigrate the work already taking place in the industry, but, at present, it’s a far cry from what we see in the best science fiction, and we should avoid using AI marketing hype to describe something that is just a complex algorithm.