Duplicity in the Duplex?

Bhavik Naik over at Bhavik’s Blurs introduced me to ‘Google Duplex‘ the other day, in a post entitled ‘Is Google taking it too far?‘ Here’s a four minute video demonstration of Google Duplex in action:

Pretty impressive, no?

As well as impressive, I also find it rather worrying… at present, this artificial intelligence widget can only — apparently at least — make appointments: but it’s not a huge step from there to being able to impersonate a salesman. Many (if not all) salesmen work to a script, one that is designed to keep the conversation on track. (I speak from experience: in a previous life I used to sell insurance.) Such activity is not so very far removed from the script required to call and make an appointment, as Google Duplex is shown doing here.

ZDNet recently claimed that this technology has beaten the Turing Test, a thought experiment in which a machine strives to prove itself indistinguishable from a human. But I disagree: although we’re pre-warned, I think it comes across pretty clearly to those listening in that the caller isn’t a real one.

In the clip above Google obviously has an agenda: it’s effectively making a sales pitch for its technology (the guy giving the presentation is Sundar Pichai, Google’s CEO). So I think that what we’re shown has to be treated with a fair amount of suspicion (unlike the behaviour of the sycophants in that audience!). For one thing, we can’t be sure what the people who were called were thinking; their priority was to close the sale, and in that situation you’re not going to risk being impolite to your caller, even if you suspect there may be something odd about the call. For another thing, both sides are working from a script — any serious deviation from it causes problems (as we saw in the second call in the clip above).

But even so, there are many issues with this. One big problem is how easily this could be scaled up, at little extra cost. At a stroke, telemarketers around the world could be out of work. (Don’t get me wrong: I hate telemarketing, but I can still feel sympathy for the poor folks who do it to earn a paypacket, perhaps even while hating the job themselves.) Worse: these people could be replaced by a system that can potentially make far more calls in a day than even the largest telemarketing team.

Have you ever received a telephone call from someone pretending to be from Microsoft? I have, several times. I know enough to know that it’s a scam (Microsoft would never call little old me), and to terminate the call. But some folks don’t know enough, and it’s those people at whom such calls are aimed. Some months ago, my mother was caught out by exactly this: and when I found out that she had been duped, it was time to retire her old computer and replace it, just in case she’d unwittingly compromised access to anything important.

Scammers like this are running a type of ‘numbers game’, where every ‘no’ gets you closer to a ‘yes’; and when you can replace the humans making the calls with machines, the numbers game just got skewed in favour of the scammers.

I don’t think that the ethical question here can be overstated. Some may dismiss Duplex as just another tool, and may claim that tools cannot be inherently bad (especially those who lobby in favour of guns). Sadly, technology is all too often hijacked by those with bad intentions. We, as a society, need to realise what may be just around the corner, and discuss how best to move forward if we’re not to be caught unawares.

Maybe the Turing Test hasn’t yet been passed, but I think that the lid may have just been lifted from Pandora’s box.