AI and Humanity: Can Technology Level the Playing Field for Equality in Society?
Summary Notes of Event
Event Description:
Hosted by the King's Think Tank
19/03/19 @ King's College London
Speakers: Olivier Thereaux (Open Data Institute) and Mark Coté (King’s College London).
Questions:
​
-
How do we remove/eliminate the risk of biases and discriminating algorithms in AI, e.g. on the basis of race, gender, and socioeconomic status?
-
How are AI/technologies levelling the playing field between the generations (baby boomers, Gen X, Millennials)?
-
If we assume that job displacement will occur due to automation and AI, what advice do you have for the public and private sectors? How should they best prepare?
-
How do we prevent data from being siloed and a state being created where the AI market is only shared by a small number of organisations?
Q1) Eliminating Bias:
Olivier’s reply:
-
We need biases in AI, otherwise they don’t do very much – so biases are ok. It’s unintended biases that we should watch out for and are not OK in the cases of race, gender, socioeconomic status etc.
-
Need to (1) understand the bias present in AI and (2) how to mitigate unintended biases.
-
It’s dangerously wrong to allocate responsibility to a machine/computer/AI – we shouldn’t anthropomorphise it, we need to get the design right. For that to happen, we need to understand and deal with issues on a human level first.
Mark’s reply:
-
There’s a legal dimension, e.g. illegal to discriminate based on X, Y, Z, but it’s also important for public to understand machine learning and be relatively fluent in it.
-
This would allow people to understand how biases emerge and understand how adjusting data sets and algorithms leads to varying outcomes.
-
-
Can’t use AI to eliminate discrimination if society and past data is riddled with discrimination
-
E.g. use of AI in recruitment based on past practice can propagate discrimination.
-
Q2) Levelling the generational playing field:
Mark:
-
Not sure it would.
-
Olivier:
-
Different generations have different attitudes to data and privacy and so will be affected differently, though this is unlikely to level the playing field.
-
E.g. millennials are most naïve and open, growing up with social media and putting everything about them “out there”
-
Gen Z are more private, perhaps learning the lessons from millennials
-
Gen X have always been more private as they were older during the birth of social media and reluctant to share.
-
Baby boomers too old to engage initially.
-
Q3) Countering Work Automation:
Olivier:
-
This is a big assumption to make, since estimates vary wildly and no one really knows how many jobs will disappear. Jobs will be lost but will also be created.
-
But, the typical response by most governments has been (1) ensure they have the right infrastructure and increase flows of data (2) train more data scientists (3) reskill workers.
-
Issue with (3) is that we don’t know what industries are safe from automation yet, so we don’t know what to reskill people into yet.
-
We should definitely wargame and prepare for changes across industries, (3) is probably the best safeguard.
-
Mark:
-
We should reconceive AI as a narrowapplication. It does one thing very well; it has a superhuman performance e.g. Alphazero, but isn’t as good at other tasks.
-
Many jobs would still need humans to do the rest of the work.
Olivier:
-
A distinction needs to be made between automation and augmentation.
-
We’ve used horses in farms historically and then we used machines in manufacturing and now we’re using further iterations of this in other industries. This augments the worker, doesn’t replace them.
-
-
Tech so far has mostly brought augmentation.
-
White collar jobs are at risk e.g. junior lawyers, as well as menial tasks.
-
Centaur model is still possible, for certain things, and would lead to a healthy future if implemented.
Mark:
-
We will see a changing of roles and an improved capacity to work in an interdisciplinary environment. Key is to participate in augmentation.
Q4) Preventing Data Silos:
Olivier:
-
There are monopolies forming, e.g. FAGAM (Facebook, Apple, Google, Amazon, Microsoft).
-
But, data is non-rivalrous: creates more, or doesn’t lose, value when shared.
-
If we allow other industries to share data and make use of it then we have value creation and then no need to break up monopolies. If the data is kept by a monopoly then the urge to break up the companies will be greater.
-
Mark:
-
The value of data increases and aggregates the more it’s shared, when going through various processes and mixed with other data.
-
It’s never exhausted over time, so we should be optimistic about the value of data and its possible uses.
-
But, FAGAM + Tencent + Baidu etc. dohave a disproportionate control over data.
-
We currently get the tip of the value that our data has when we give it to them, but they reap the rest and aggregate that value with everyone else’s data.
-
Mark’s Q: if data is non-rivalrous then why is there still a narrow set of applications for our data?
Olivier:
-
GDPR gives us certain rights (1) accessibility (to see data) (2) to be forgotten (to delete data) (3) portability (to move data, e.g. to a competitor).
-
(3) is not well known, the least used, and the least enforced.
-