“No don’t get the fried stuff” “why?” “because we don’t want
it” “but last time you saw the chips packet you guys hammered it down and finished
it. You guys eat it.” “ya, exactly. We eat it. Doesn’t mean we want it. We don’t
want it. We don’t want to eat it and so please don’t get it home. Right now we
are enjoying eating home cooked food and good food through the day. We will eat
fruits. Whatever. But please don’t get the fried stuff.”
Jasmine is the ever care giver. She sees what her babies eat
and takes that as the cue to what they really want to eat. For years her
strategy isnot to really ask us. We always give the wrong answers, she
contends. She sees how we react and then she figures. By now she knows under
which circumstances Jispa will eat more food. She knows that when she serves
hot rotis I eat a few more. Which is a problem. I don’t want to eat more. I eat
more because she is making them. It’s a circular thing.
Nothing worries Jasmine more than seeing an empty bowl of
food after dinner. “Oh. You want more, should I make something.” “NO.. I don’t want
more. I was just trying to finish the food you made. I thought instead of
throwing it, since its less, I thought I will eat it and finish it.” So now when
I finish the bowl I make it a point to first ask everyone if someone wants and
state that I don’t want and then with the mandatory disclaimers, I finish the
bowl. I don’t want it, its extra but still I am having it. Then it works.
While writing this, I realised an interesting problem of artificial
intelligence. Hmmm. This is the same problem that Ai would face while decoding
human behaviour. AI is very similar to Jasmine. It seeks to capture exhibited
behaviour and response to stimuli. It doesn’t ask the consumer what he/she
wants. Its simply sees behaviour / stimuli and adjusts accordingly. This way it understands you better than what
you do of yourself.
This is the hallowed promise of AI. But in this is also the curse
of AI. It will seek only exhibited behaviour and not consider motivation behind
the behaviour. This could be the achilles heel of the Ai as we implement it.
While doing one AI project, I realised that one of the problems that AI had was
that basis behaviour we would give out stimuli and basis the stimuli we would
see further behaviour. This seemed to me a self-fulfilling prophesy. The whole problem
would be that if we are stuck in a behaviour and stimuli that is of our doing –
we will miss if there is a behaviour outside of the proverbial well. That’s why
I insisted on inserting trojans - Randomly generated stimuli even if it against
the data of the behaviour. This way we would also be able to catch a random
trend starting. So just because a type of behaviour was not seen so far doesn’t
mean it doesn’t get captured.
It’s a bit tricky to understand. To simplify – the AI may
start seeing that every time Jasmine makes hot roti, they eat more. So, it
churns out more and more hot rotis not knowing that possibly sometimes or every
once in a while, Jasmine should churn out Hot Rice to see what happens. Are bored
of hot rotis or a new trend towards brown rice is coming in that the AI doesn’t
catch as the current data continues to show upticks. It’s what I would call a ‘staying
in the well’ syndrome.
The other thing I realise while writing this blog is that the
AI must consider human motivation for behaviour. So, while I am eating more
rotis, is it that I really want to eat more just because they are hot. May be a
deeper insight into the behaviour of me eating more can be taken in. Possibly I
love eating rotis but more importantly I want Jasmine to eat along with me. I don’t
mind eating a roti less. Understanding this is a crucial dimension for AI going
forward.
Simply blindly following correlation over causation will
only take you this far. You easily find yourself on top of the curve that has
flattened and leave you with no growth opportunity. Combination of trojans and insights
and research can add into the mix so as to code for growth that is break
through instead of just paying lip service to ‘I have implemented AI’.
Most CEO’s do not understand this and simply expect an AI
engine to ingest more variables. They like saying we have AI and that we have
ingested 5000 points of data. It sounds very good and it a good lie to perpetrate
but it’s just that – a lie. AI would need some serious soul searching on behalf
of the coders and marketing nut cases to make it work. I shudder to think of
what would happen over a period as I exited my previous company where I created
this juggernaut.
Sometimes I feel like a physicist who cracks the code for
energy is dismayed to see it being used for weapons of mass destruction. I create
an AI engine to see the top use it for random things. They just don’t understand
it. Then I connect it to a mass communication system only to see that they will
use it to think through wrong conclusions and send wrong communication to a
wrong audience and overdo to kill it. Entropy is real.
So coming back – Jasmine is enjoying the fact that during
this lock down she is able to feed us hot rotis and we are fighting back by
stating upfront exactly how many we will eat. This way its in control. And both
win. Plus I sit to eat a little late – I keep telling her that I will sit only
delaying it enough so that when she done making my hot rotis I will have at the
table eating her dinner with me.
I really care about talking to her more than I care about
getting fatter. Wouldn’t you agree?
