Page 146 - ITU Journal - ICT Discoveries - Volume 1, No. 2, December 2018 - Second special issue on Data for Good

,78 -2851$/ ,&7 'LVFRYHULHV 9RO 'HFHPEHU

goods of their fellow citizens or incompetent, or 4. CHALLENGES AND IMPEDIMENTS TO
both. Instead, a human AI would be a system where DESIGNING A HUMAN AI
it would be difficult for an elected representative to
claim credit or assign blame out of hot air, because This sketch of a human AI has left out many
citizens would say “Really”? Show me the data!” Of challenges and questions, the biggest of which we
course, it would allow well-meaning politicians to can only briefly discuss.
do a much better job but it would be a system where
citizens could fight bad politicians armed with First, some of the examples mentioned above are
better data, and make better decisions for and by voluntarily contentious because there is no
themselves. consensus on them. For example, many people
around the globe still think that spanking a child is
It is neither an “Orwellian” vision where citizens’ good for him or her; many people support the death
actions would be digitally monitored and rated in penalty and torture; many people are skeptical or in
real time all the time. This risk merits more denial about climate change. Some people still insist
attention than is possible in a few pages, but a few the earth is flat. More people believe tax cuts for
points can be noted. First, all societies have systems billionaires create jobs. This raises the general and
in place to influence individual behaviors in ways fundamental question of what, whether and how we
they deem desirable, from taxes to laws via credit do, could, or should learn, individually and
scores. Second, the focus of human AI is not collectively, and how to come to a consensus and
individual actions; it is about instilling a culture and reach a compromise on major societal issues.
setting up the necessary systems and standards to
improve collective actions and decisions. Some people will not figure out why they are no
longer invited to parties, or will not ever be able to
For this to work, there first needs to be a general adjust and stop talking loudly all the time. We are
agreement that decisions and outcomes ought to be not fully rationale beings either, and lots of other
evaluated on the basis of data, which is for now used considerations get in the way. We (should) all know
as synonymous with facts. It may not be easy to that getting that third piece of apple pie is bad for
agree on the features and factors of “good results”, us, and yet many people do because humans tend to
but at least we should agree to assess them on the value the present more, especially after a glass of
basis of facts. We need not have a preconceived good wine. Others choose to disregard or selectively
agreement on what level of income inequality is pick data (here still synonymous with facts). They
desirable, but we should start by agreeing that and may believe that climate change is a hoax because
how inequality will be measured. From there we can they are told so by sources they grew up hearing.
understand what contributes to different levels of This is especially convenient if their income
inequality, and what outcomes these differences depends on fossil fuel. Others who have
result in. A human AI requires a general agreement experienced and then imposed domestic violence
that facts should matter, in a “Northien” perspective for decades may prefer these actions being
[11], because otherwise systems cannot learn; and comforted than confronted. Members of the world’s
if they cannot learn, they cannot improve. ‘intellectual elite” discuss climate change in
conferences that require them to fly thousands of
In summary, a human AI is a human social system miles, convincing themselves that on balance the
that would apply and leverage the power of data world is better off that way.
and the “good magic” of AI, the ability to assign
credit and learn from feedback with data as key There is also evidence that facts alone do not change
inputs and outputs, to reward and reinforce people’s minds [12]. This is not new to the data era,
decisions and actions that contribute to good but there is a sense that in a world awash in data, it
results, through and feeding fact-based discussions becomes even harder for facts to be recognized and
between its members. agreed upon [13]. While so far we treated data and
facts as synonyms for simplicity, there are obviously
differences between the data we swim (or drown)
in, and facts and truths. Ours is also the world of
fake news, alternative facts, where advances in
digital imagery may soon mean we should not
believe what we see [14]. Ill-intentioned powerful

‹ ,QWHUQDWLRQDO 7HOHFRPPXQLFDWLRQ 8QLRQ