I once had a salesman trying to sell me a Neural Network (AI if you like). It was obvious that teaching it all our business rules and so on would be so time consuming it was more efficient for people to just get on and do the work. HAL 9000 it wasn't. Remember? We had the same buzz about Big Data, and Blockchain / Bitcoin, only some months ago: they also were The Solution to Save the World. AI is just the latest avatar for these technical dreams. All share a common fact: they hardly have any reality beyond PowerPoints, almost none of the promises have been / will be delivered. Meanwhile, beyond the journalists and public being tricked, the CEOs and IT manager themselves attend conferences, watch marvellous PowerPoints, dreaming about those miraculous new technologies, and spending millions of real pounds, dollars and euros to hire AI specialists (but hardly any Data Scientists anymore...).

So I'd say this is mainly marketing bullshit, very well sold.

I call this a misadventured attempt at misdirection.

As an engineer I’ve long been skeptical of technology for technology’s sake and as we don’t know what consciousness is and hence sentience then I am doubly jaundiced about AI. I am also concerned at the ever increasing of un-needed layers of complexity we are piling into how we get things done in society. Sure, I understand if you do online banking you need a certain level of security which requires a certain level of complexity but now we are being told that we need to get rid of cash transactions which just cements in a certain type of complexity.

In the industry I work in the processes have been entirely replaced by server based systems with their own problems. These problems are always complex but always the same, reboot the driver, etc. The thing is, a huge amount of complexity has been introduced for very little gain. We used to use a particular technology that has been almost supplanted by ‘digital’ but it costs a very large amount of money for ‘digital’ to reach the same quality as the previous technology. However, most people involved, either technical or non-technical, generally express reactions ranging from denial to bewilderment when I question whether we have gained anything.

When I read it a long time ago, I thought it was no big deal. In retrospect, I still think it's nothing to brag about. There are much better books in Academia than this one.

Why I selected an AI manager over a human:

My AI manager doesn't smoke and isn't a malicious gossip.
My AI manager assesses my work objectively.
My AI manager learns from me but isn't allowed to steal my work.
My AI manager is aware of its own limitations.
My AI manager isn't illiterate.
My AI manager isn't thick.
My AI manager can't skive off early.
My AI manager can't avoid responsibility.
My AI manager can actually understand and evaluate a counter-argument on its merits.
My AI manager can't lie.
My AI manager can't look after its mates.
My AI manager can't harass me or ignore my complaint if I'm being harassed.
My AI manager isn't human and I don't have to pretend it is, in short I don't have to humour it.

If you go by this book, and don’t believe all of the above, I'm guessing you’re on your way to know about AI. Incidentally, none of the above four would be true ....for starters. But that’s OK, because the only point on the list would be 'My AI manager doesn't need me, because it has other AIs, so I have no work, no income'.

Books like these, remind me of the unmitigated optimism displayed about automated language translation that overtook the computer science community. More than fifty years later we are no closer to achieving automated language translation than we were back then.

Translating a piece of text from English to French or Italian can be done relatively easily by a person skilled in the source and object languages, but it is beyond the capabilities of computers. Why? Because somebody has to write a program to do the translation and take into account the idiosyncrasies of the languages involved. It can't be done by dictionary lookups.

Artificial intelligence (which is really machine learning) is in the same bind. While we can teach robots to do simple tasks like painting a car or welding two pieces of metal together, life consists of far more complex activities that can't always be reduced to simple binary options. How do you program a driverless car to crash into a wall and endanger the occupants rather than run over a couple of pedestrians?

Journalists should be less ready to accept the pap that is produced by technology companies promoting their futuristic products.

As far as I can see Machine Learning is the equivalent of going in to B&Q and being told by the enthusiastic sales rep that the washing machine you are looking at is very popular (and therefore you should buy it too). Through clenched teeth I generally growl "That doesn't mean I think it is the best washing machine." Following the herd is not my bag; there are enormous problems down the line: the circular argument of how people make choices is strengthening its grip as real-time information (likes and dislikes) accelerate across social media networks. A ghastly law of averages and gambling emerges — a polarising exponential effect to choice making: if my friend likes it, if 1000s of other people like it, if millions like it...then it must be likeable and therefore I must like it/want it. It can be anything from a washing machine to a football team to voting in an election. Individuals are losing autonomy in their thinking and decision making; they follow the herd bumping up against the limits of their own echo chamber; they run at speed from the white noise of too much information.

This isn't intelligence (artificial or real). This is lemmings following each other over a cliff.

Bishop's book is an oldie, but it's still one of the best treatments on AI around. Even today. ( )