Pi, designed to be a personal chatbot that can develop into virtual companionship over time as it learns from your personal interactions, launched today. The team behind Pi is an all-star cast: Ex-DeepMind/Google, Reid Hoffman, OpenAI board member, etc. The team has quickly raised in excess of $225 million. Despite everything going for it, on initial use Pi is deeply “underwhelming,” the AI wet blanket, and this has implications for what comes next. Let’s explore:
There has been speculation that there will be “one AI to rule them all” and a large swath of silicon valley is chasing after this notion. Hundreds of millions have been raised by similar concepts (character.ai for example). Pi’s launch today has made it clear that similar to the automobiles we drive, AI bots are tools (as of now) and there will be many opportunities for varied brands and functions to exist within the space. These will serve niches as varied as the difference between a dump truck and a Ferrari convertible. As a result, it will be difficult for a single company to form a monopoly.
With respect to Pi’s launch today, perhaps there is a large segment of the market that wants a completely vanilla friendship that is confidential and always available. I can certainly see this as an avenue. Reid Hoffman is pretty smart so likely there must be a large market for this concept.
However, for me personally, Pi was bland and served no utility whatsoever. First I tried to see if it could help me make a restaurant reservation. No. Then I pushed the boundaries in a NSFW Replika direction. No bueno. Trying a new tact, I thought I would test a difficult scenario I’ve been helping a friend with where they feel undermined and demeaned at work to the point where the situation is significantly affecting their quality of life. Pi’s advice was to talk about it and if that doesn’t work, “Perhaps the best thing you can do is keep showing up… Over time, maybe they’ll start to see that you’re reliable and trustworthy”. It’s hard for me to imagine this is the caliber of response any coach, therapist, mentor, or close friend would give.
It was in this realization, that Pi cannot perform well as a coach, therapist, mentor, or close friend (or personal assistant or romantic partner), that we can see the granularity and potential size of the market offerings. At the very least I see a market for the following:
The do things for me AI, or multiple function or task-specific ‘do things’ assistants (future post)
The personal connection AI (some point - or multiple points - on the spectrum from the vanilla acquaintance Pi offers today to a “Her” style romantic relationship)
Mentor/Coach/Therapist: Again these may all be from a unified system of thinking or might be multiple differentiated offerings.
In all of the examples mentioned, there is obviously going to be a desire to tune or select a style (there are multiple commonly accepted approaches to trauma therapy, for example). At least as of now, there is one option for Pi and you cannot make alterations.
If you subscribe to the saying, “you are the average of the five people you spend the most time with,” and believe this has transitive properties to chatbots, I wouldn’t recommend Pi for the lineup. Or, at the very least, I’d keep searching to raise the bar with the next four.