AI (artificial intelligence) is the current hot topic (2020s), but is it a future we must all embrace, and if so what does it involve and deliver? The trouble with such technological evolutions is that promotional marketing of a deep technology inevitably involves misunderstanding and misinterpretation. The levels vary: Y2K was little better that a scam; IoT is genuine, but the technology to do it was available decades before it was introduced, as something brand new, and it still has a long way to go; and the world wide web was transformational and rightly sold as so. So, where does AI lie?
The answer is it is certainly transformational, but it isn’t intelligence. It is almost certain to be misunderstood as such unless enough knowledgeable sources make efforts to slow the AI marketing juggernaut.
What is AI? What isn’t it?
Comparing the interface a machine sees with the world with that of a human is a good way to answer these two questions. The human, or animal for that matter, is truly intelligent, and it is this comparison that highlights what really qualifies as intelligence. In the following paragraphs machine and human ability are defined and compared against an accepted definition of intelligence.
A human has all these mental capacities. AI only has to arguable degrees: communication, learning and memory formation, action planning, and problem solving. The day that a machine can truly think like a human, the secret of life will have been discovered. AI cannot think. It simple collects vast amounts of data from its sensors and applies the mathematics of statistics and probability to the data to make decisions and, if dictated by its programmed algorithm, adjust its decision making limits. The rules by which AI makes decisions are fixed, mathematical, and programmed by a human being. The difference between embedded computing (microcontrollers) and AI is that traditional computing uses fixed rules that are always applied (usually on a very limited data set), but AI allows for statistical analysis of significant bodies of changing data from which rules may change according to the mathematical rules of probability and statistics, but it is human intelligence that has set fixed probability and statistical algorithms for the AI.
The essence of AI is that it allows for the analysis of huge data sets in order to assess the probability that something is true or false, whereas embedded microcontroller systems essentially work on a binary true of false decision making system.
The hardware is unchanged except in that greater computing resources are required for AI. In other words, there have been no special or new hardware technologies incorporated with the advent of AI.
A practical example may assist with comprehension of the above. The Electronics Machine has developed an innovative device that detects drowning. It connects to swimming goggles. When people drown (apologies for the morbidity) they adopt a ‘climbing the ladder’ position that is vertical with the head angled backwards. The nature of drowning is such that the victim bobs up and down in the water, so detection is also concerned with water monitoring. At the time of development the problems were solved with binary decisions about the angle of the head and submersion in water, but this example is well suited to AI. If the complete data set of all angle measurements and water submersion statistics were taken and from these the probability of a given set of results signifying drowning determined, a wholly more satisfactory solution less prone to false positives would result. If a front crawl swimming session suddenly had stats like those above inserted in the data set, the likelihood of an incident having occurred would be much higher than the immediate assessment of a binary detection. What’s more, it is likely that other interesting things could be detected.
There’s an engaging parallel with a profound statement in literature: Great Expectations (Charles Dickens). A scene involves Pip’s sister being attacked and the ineffectual visit by the police afterwards. Paraphrasing, the policemen tried very hard to fit the circumstances to their ideas rather than using the circumstances to give them ideas. The parallel exists in the realisation that embedded solutions set fixed trigger limits (ideas) that are hoped to detect a condition in the data (circumstances), whereas AI analyses all/most of the data (circumstances) and seeks to find its trigger limits therefrom (ideas).
AI is a misnomer. It is not artificial intelligence. AI is nevertheless a very much more advanced and intelligent way for machines to interface with the world. It is the future, but it would be better to rename it appropriately and not mislead millions of people.
AI is a cutting edge topic at the moment, and a company that can claim its use will gain kudos just by being able to include it in its sales pitch. Such a claim will obviously be hollow, but the point bears making.
So, where is AI truly of value? The answer is in cases where a view of sensor data on mass allows probabilistic judgements to made. AI projects will therefore have much better coupling with the objective world and give the impression that human intelligence lies behind them. This brings in a need for powerful computers and advanced signal processing (DSP) at source. An example would be any application processing digital camera feed.
Fortunately, there is a lot of help via substantial electronics companies, e.g., Renesas. Bespoke software to model the statistics and other tools and most particularly a service that offers appropriate hardware for the design.
We are more than happy to assist with such designs and turn dreams to reality.
Get in Contact Today to start your AI Development Journey
