A lot of interface level AI is about helping people speed up interactions. Like this example from Google.
However with more complicated examples, how do you know that the recommendations you are getting have your best interests at heart, rather than those of the companies advertisers, sponsors.
For instance, if you ask Alexa to add something to your shopping list, how do you know that they're going to add the cheapest version available? Or the version they'll make the most money off?
For AI Agents to work, you need to have a huge amount of trust in the brand. Or at least be comfortable that the premium you may get charged as a result is worth the fractional increase in speed.
Of course this isn't limited to AI agents. When I buy groceries using an online app, I'm trusting that they don't add a huge mark-up over their in-person store. However it's much easier to comparison shop through a browser than a CUI interface with a single agent.
The big tech companies failed in their walled garden approach to the Internet, thanks largely to the browser and the open web. Apps have limited that flexibility somewhat. Conversations Interfaces even more so.
One of the interesting thing about software agents is understanding what they're optimising for. For instance my sat-nav generally optimises around speed, but I can set it for fuel efficiency. However I'd actually prefer to optimise it around the pleasantness of the journey.
For me, pleasantness would be a combination of...
1. Nice views
2. Not too much traffic
3. Not stopping and starting too much
And I'd LOVE a sat-nav that suggested a route that took me past a lovely rest area, rather than a corporate service station.
1. Nice views
2. Not too much traffic
3. Not stopping and starting too much
And I'd LOVE a sat-nav that suggested a route that took me past a lovely rest area, rather than a corporate service station.
Many years ago I was involved in a project to surface the information a large media/tech company had on you, and how this was used to tailor recommendations. So you could essentially look at a "data layer" sitting behind the recommendations.
Such a shame in never came off as I reckon t would have a) been ground breaking and b) set a standard for others to follow.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
