Google started its annual I/O conference yesterday. It was packed full of AI applications showing that Google really is going ‘AI First.’ It even changed its research arm name from Google Research to Google AI.
But perhaps the most draw dropping demonstration was when Google used an AI to make calls and book appointments. I strongly recommend you spend 5 minutes and watch this.
Now this is obviously cherry picking. It would be great to hear conversations that didn’t go so well from the perspective of the AI but I think we can saw that Alan Turing would be jumping up and down for joy at this.
Suffice it to say, if an AI can do this, I would certainly be happy for it to make calls on my behalf. The utility of not having to have such conversations would be immense. Indeed, I use online options where ever possible just to avoid these.
However, as an economist, I need to consider the equilibrium. If you lower the cost of making calls, people will make more of them. For appointments, that may not be an issue but remember, there is a person on the other side of that call. At the point where the AI is good but not perfect, that may turn out to be very, very annoying. For this reason and perhaps others as well, many are advocating that you are informed if you are talking to a robot or not — as a public policy regulation. Their intent is against malicious AI that may harm the receiver.
We need to weigh this against potential costs. Will the calls be less effective if you know it is a machine first? If they aren’t good, then people will hang up on them. Moreover, if it is a standard announcement, there is no hope for Google — who may have a good AI — to distinguish itself from others that don’t. That said, while my instinct was against regulation, these costs, at first glance, do not seem that low and could well improve innovation in ways that benefit both sides of a call. Which machine is calling could be identified to so that receivers can distinguish the good from the bad.
The bigger concern is spam. Every other time we have given robots the ability to make calls it has ended up terribly. At the moment, generic robo calls are flooding phone networks. I no longer pick up from unknown numbers but that doesn’t stop them clogging up my voice mail.
My belief is that Google should use this new development to get ahead of the issue. At the moment, we have a human way of dealing with calls. You see a number and you decide to pick up. This is craziness. Instead, your phone should be able to screen calls for you. For starters, Google could have a signal that is sent with its calls that it is recognised by phones (or phone networks). Users can then pre-program whether they will accept such calls or not. This could then be used for robocalls as well and also for legitimate callers. I am not sure how to do this technically but this is the sort of thing that strikes me as very solvable. Do not call lists are old school and don’t work. It is time for the technology to improve.
Second, it seems to me that the era of machines talking to humans is likely to be a very short lived one. Yes, Google’s demonstration is impressive. But really. What should happen is that the hairdresser or restaurant has its own machine answering the call — we could call it an ‘answering machine.’ They get the signal that it is a machine calling and they use a machine to answer. There is no conversation as we would recognise it. No need for ‘hmms’ etc. Instead, the appointment would be negotiated. The bottom line is that the issues everyone is concerned about today really don’t seem like the real issues.
The missing ingredient in all this is a standard for identifying machine callers (potentially quite deeply). Along with this, security elements could be backed in so that a machines could not be tricked into revealing too much. For instance, one could imagine that a celebrity’s schedule could be discovered through such calls.
3 Replies to “Should we let AI just make calls?”
If the person/business you are calling needs to make an effort to install an ‘answering machine’ with knowledge of the booking calendar then they could have an online booking system which would remove the need for the call in the first place.
Perhaps the upshot of this will be to fill the phone network with appointment booking spam, and force small businesses to get websites?
I’m also looking forward to the hilarious mistakes that will ensue. Currently, my ‘smart’ email extracts my rail journey times when it notices my ticket confirmation messages, but assumes I live somewhere six time-zones earlier than I am, so all my ‘appointments’ are in the small hours of the morning.
I also have ‘Automatic’ processing of receipts for expenses, but I spend just as much time going through correcting its crazy decisions/categories as it would take me to do the expenses in a spreadsheet myself. eg. It frequently just grabs a random number off the receipt and assuming that is the ‘total’. What’s worse, I’m pretty sure that system is mostly done by humans.
Oh, and the ‘smart’ substitutions to my internet shopping basket last night. ‘Sorry, we don’t sell individual apples any more. Shall I substitute pre-packed?’ …and then getting 36 apples instead of 6 because it had substituted one pack for each apple.
These can be dismissed as teething problems, but as a software developer who has had to patch systems that have been ‘improved’ continuously for 20 years and still do unexpected things all the time I’m not holding out any hope. And that’s without people deliberately trying to confuse the system for fun.
‘I’d like a table for 7’
The minute I watched this skit when it was shown on Google i/o live, I immediately wondered regarding the Turing test. Surely we are not interested in ‘experts’ or being ‘limited to using the responses to written questions’ to make the determination’ in order to distinguish computer from human? My elderly mother is, like many other elderly people, and is increasingly as I am, bombarded throughout the day by call centre operatives aiming to guide her towards new windows, free gifts, hearing aides, or insurance. Its changed the relation she has with the phone [she is not connected but does play solitaire on her iPAD). For all intensive purposes such a system would fool her, extract information, possibly with the sinister potentials of using ‘long con’ tricks and tactics that at least seduce her in a far more pervasive and persuasive manner than the Indian or Filipino script being read over the phone – that is regardless of the localisation accent voice training these operatives receive. Like my proflirating email box, where it is impossible now to separate chaff from the wheat (even though spam filters out penis extension and foreign offers of love), we face the extinction, at least, of this means of communication.