When will a regular person be able to use AI to answer phone calls for them?
41
Ṁ3264
2030
8%
Before 2024-10-01
20%
Before 2025
69%
Before 2026
71%
Before 2027
79%
Before 2028
80%
Before 2029
94%
Before 2030
Resolved
NO
Before 2024-07-01

The market will resolve positively when a product or service appears on the market which will allow regular users to use AI to answer phone calls for them, acting as their secretary. The AI should be able to have a back-and-force conversation with the caller and to make at least some decisions for the user. E.g. receive a call that some event is cancelled and remove it from the user's calendar.

Just refusing the spam calls is not enough, but if the AI has a conversation with the caller and decides whether the call is spam or not based on that conversation and taking into account user's preferences, then it will qualify.

This service could be an app, a phone feature or an online service like Google Voice. An exclusive feature of a particular mobile operator doesn’t count because I likely wouldn’t be able to try it out to see if it qualifies.

I do not bet on my own questions.

Clarification from 2024-07-01: the phone-answering AI should be smart enough to make at least very basic common sense decisions, like "No, my employer is not a plumber, so I won't make an appointment to fix your faucet"

Get Ṁ1,000 play money
Sort by:

@traders I'm about to resolve "Before 2024-07-01" as NO, unless anyone has an example of a service fitting these criteria.

It sounds like AI Lucy acts basically like a slightly smarter voice mail. It seems like it doesn't fulfill the following requirement from the description: "The AI should be able to [...] make at least some decisions for the user."

Their website list the following things that Lucy can do:

  • Share your appointment booking link

  • Share your pricing information

  • Share your website link

  • Share your email address

  • Take a message

Elsewhere they claim that she can make and cancel appointments, but I don't think she can answer something like "No, my employer is not a plumber, so I won't make an appointment to fix your faucet".

It sounds as though they may be able to arrange an AI to selectively screen for inappropriate appointments. (It’s also possible the basic LLM model does this—albeit they don’t explicitly advertise it.)

If you’re after a more advanced AI than Lucy, you’re in luck. Curious Thing offers other products like Alex for more complex inbound call handling, and Sam and Lou for outbound AI calls, please request demo to learn more.

Alex looks like a B2B product for phone support, which again doesn't quite fit the requirements.

It is possible that Lucy is smart enough to make some decisions. Unfortunately it's not available in my country, so I can't try her. I also didn't manage to find any reviews, which makes me a little bit suspicious.

I'll hold off resolving "Before 2024-07-01" for a couple of weeks in case somebody manages to try Lucy out or I find some reviews.

Before 2024-10-01

If some it doesn't happen, will this be resolved on that date or in 2030?

@intiluha Each date resolves to NO as soon as it passes and the service doesn’t materialize. Then when the product appears on the market, all remaining options resolve to YES.

Since the current estimations are super optimistic, I'll add a couple options in the more immediate future.

If it appears in 2024, do all options resolve to YES?

bought Ṁ20 Before 2025 YES
bought Ṁ30 Before 2030 YES

@OlegEterevsky one more question: OS features shall count? Such as : service rolled out as a part of the next iOS or android release -> some new phones get it, some old phones don't get it

@firstuserhere Absolutely. I count it as a “phone feature” that I mentioned in the description.

@OlegEterevsky OK great. I wonder if you also want a qualitative lower-bar - so that it should actually be a usable feature.

I get prompts from google to let their AI answer spam calls on my behalf, I will try to use it the next time and see how good it is

@firstuserhere Yes, I mentioned that ot has to be able to have a back-and-force. I don’t want a glorified spam filter to count. Mind if I explicitly add to the description that just refusing spam calls is not enough?

@OlegEterevsky yes, that'd be great

@OlegEterevsky have you used https://www.retellai.com/

If you haven't, I recommend that you do!

@firstuserhere just scroll down and test a voice call

@firstuserhere I updated the part about the capabilities.

I didn’t know about retellai, will try it a bit later today.

@OlegEterevsky Great, retell ai is targeted exactly towards this. I did a test call and honestly, it was good at detecting the umms and uhh etc and actually responded fairly nicely

@mattyb No, the AI has to answer the calls for you, not call on your behalf

@firstuserhere I looked at https://www.retellai.com/ and it seems like a B2B service, that provides the capabilities, but doesn't actually implement the end-user product. Is that correct? In that case, it doesn't resolve this market.

@OlegEterevsky you can try it out for the qualitative aspect. Booking the call is free

@firstuserhere Oh, I have no doubt that GPT4 + Whisper are capable of it.

My question was specifically about whether somebody will release this as a product. Like, for example, Google Duplex was demoed already in 2018, but it wasn't ultimately released as a product.

@OlegEterevsky I am still a bit unsure about what level of quality/usability is enough for this to resolve yes. Technically, the AI telling me nonsense and me repeating what I want is a "back-and-forth conversation". I agree that something that works specifically for business calls at least say 30% of the time looks pretty much within reach. On the other hand something highly reliable across a wide range of contexts seems much more challenging...

@MartinModrak If there's an edge case like that, I'll resolve one or more dates as N/A.

But I don't think this is very likely because a) having a normal back-and-force conversation is mostly a solved problem now. You can just tell GPT4 to do it, and it work fine most of the time. More importantly, b) the resolution criteria mention that it has to be released as a product. If it's just spewing nonsense, it's very unlikely to be released.