MIT 6.S087: Foundation Models & Generative AI. ETHICS

all right let's get

started so uh today should be very fun

because we're going to talk about ethics

and regulations so first I'll provide a

kind of a very high level lecture cover

a lot and then we're going to have a

panel uh where manolis comes back uh and

you get a chance to ask a lot of

questions we can discuss this uh more in

detail right so I mean there's there's

little doubt that Foundation models and

generative AI becoming uh very impactful

as a technology right so then the

question is how do we develop this

safely and responsibly and what kind of

air do we want and what can we affect

and also should we regulated and how can

we balance Innovation and regulation is

what we kind of try to uh cover today so

I'm actually going to open up and asking

you a question who would you blame if an

AI hurt

you I think this is this is a key

question and so there's a lot of

different

Alternatives um maybe you want to right

let's say tatb does something should you

blame the company that developed it open

AI or is it perhaps in institutions like

government institutions agencies that

supposed to regulate things and protect

us is it maybe you know the AI itself is

an entity or a person is responsible

like we're you know we see AI as as an

individual with its own aims and and

desires is it the data that's been

trained on

is that the problem or is it maybe you

know the best kind of tool that we have

is the people behind it right the

stakeholders behind this AI

um and I think like anthrop

anthropomorphizing sorry

anthropomorphizing AI is very dangerous

because it makes us feel that AI has an

agency and responsibility and it can

allow stakeholders to hide behind Ai and

relieve themselves of responsibilities

uh but at the end of the day still right

even this type of AI that we're seeing

right now that's very powerful there's

always people behind the AI That's

benefiting from it right and that's the

ones that that somehow uh probably

should be kept responsible I think

that's a very it's a useful perspective

because they're they're motivated and

they have G something to gain or lose so

you can try to understand what AI is

doing but looking at the stakeholders

behind it so right if you ask CHP itself

to gbt forhead to create an of yourself

you get this you know personlike entity

that media also likes to portray which I

think is a dangerous and false

representation that allows uh the

stakeholders behind them to hide behind

the problem of the AI rather than the

problem of the stakeholders right that

wants to build it and benefit from

it okay so what are like let's take TP

for example in open eye what are the

stakeholders well you know there the

people behind the company uh uh the

leader the management Etc right we have

my picture here then we also know that

Microsoft has bought a 50% stake in

openai so then suddenly they also kind

of Stak holders and have a lot of

motivation here uh and then on top of

that right suddenly the CEO of openi is

pushed out and then he's he returns

right so this this a you know when we

talk about transparency responsibility

in terms of AI models there's a huge

lack of transparency in the stakeholders

Behind These AI models and I think it's

it's kind of a you know great challenge

to kind of hypocritical that these

people are supposed to build transparent

Ai and that's what we care about when

the organization is completely you know

non-transparent and they can't keep

their things together right and that

should make us worry this is nothing

that's not related to AI at all but the

problem already starts there right when

you try to establish who's the

stakeholders okay so

um right so we're going to start off

covering a lot of different uh topics

BAS basically the first part will be you

know jumping between topics of AI and

ethics and then we're going to say we're

going we're going to say like well these

are complicated issues and then we're

going to see how government and

institutions try to regulate and address

this issu and then we're going to talk

about potential future

directions and I think also I don't want

to like pretend there's a mainstream

perspective on these issues right these

are very nuance and complicated I want

to hide behind some some uh you know

mainstream or say that these are you

know solved solved problems and there's

clear answers so just to be completely

you know transparent uh this is you know

I'm sharing these uh thoughts and

perspectives H and really it's just to

get you guys to think about these

problems also just make you ask more

questions really uh so I'm going to

cover a lot of different things and jump

back and forth not going to cover

everything just going to try to make you

guys think and and be excited and try to

make up your own opinion right I don't

want to lect you and tell what's good or

bad and moral that's for up for you I

guess to decide I want to provide you

with certain aspect and problems that

we're seeing in

AI okay so before we jump in I just want

to emphasize certain perspectives so one

perspective I think is important is that

is AI categorically different from other

Technologies or other problem areas that

we've been seeing in our history right

maybe AI is just part of a type of

development or Evolution I've been

seeing for a long time maybe AI is just

you know part of a Continuum rather than

something that's very different and

maybe AI is not the problem in this

trajectory maybe we went wrong somewhere

earlier maybe when we started using

electricity maybe that's where we really

went off right and did something bad so

also like is AI really that different

from what we seen before I think it's an

important uh perspective and and

actually is this not where a real

challenge it's not AI but something else

okay another important point I think is

that ethics is a lot about how things

should be how the world should work

right philosophy and ethics love to talk

about the shoulds I think it's a big

difference between how the world should

work and how the work how the world

actually works right so I think this for

me I think it's less utility in building

up a Utopia of how we would like the

world to work I think it's more

interesting talk about how we think the

world will work uh and you know realize

there's a difference

here okay and last well I mean so these

are you know real problems that are uh

high stake problems right and now

there's definitely you know using these

Technologies uh to boost their nation's

military security and it's becoming an

arms race and I think it's also you know

that makes the stakes higher and certain

things become like a necessity to

protect your nation and typically when

there's an arms race it's a lot of you

know there much less should but it's a

lot of competition

uh and things move very very quickly um

yeah so I think that's also you good to

keep in mind that these things are not

just affecting ourselves but they're

affecting this you know bigger picture

uh right everything is is fair in Love

and War and I think that's uh

historically quite

true okay so we're going to start off

walking through some different threats

that we see from

Ai and well there's a lot of them right

we have misinformation and manipulation

we have deep fakes we have privacy bias

and discrimination lack of transparency

and accountability centralization job

displacement unnatural living conditions

and the unintended unknown so let's get