INTHEBLACK February/March 2024 - Magazine - Page 22
F E AT U R E
“We don’t have ‘explainable AI’ at the moment. People are trying to work out how to interrogate
what’s happening, or to create systems that can be explained, but it means we are currently using
things that we don’t fully understand in terms of their internal operations.”
SIMON LONGSTAFF AO FCPA, THE ETHICS CENTRE
Above: Microsoft CEO
Satya Nadella (right)
greets OpenAI CEO
Sam Altman (left)
during the OpenAI
DevDay event, San
Francisco, California,
November 2023.
LISTEN
to an audio recording
of this story
22 INTHEBLACK February 2024
“We have AI that we know works, but we
can’t explain precisely why,” says Dr Simon
Longstaff AO FCPA, executive director of
The Ethics Centre.
“That’s because the way the large language
models and other systems operate means they
can be trained to produce a result. However, it’s
not like an older form of computer logic, where
you could see that A leads to B leads to C.
“We don’t have ‘explainable AI’ at the
moment,” Longstaff adds. “People are trying
to work out how to interrogate what’s
happening, or to create systems that can be
explained, but it means we are currently using
things that we don’t fully understand in terms
of their internal operations.”
Regulation has always required appropriate
levels of governance around technology. When
personally identifiable information is involved,
for example, there are regulations that dictate
what companies must and must not do.
Professor Matt Kuperholz, data scientist
with Deakin University’s Centre for AI
and the Future of Business, notes that
forward-thinking companies have already
created “ethical guardrails” to ensure their use
of these technologies is appropriate and in line
with their customers’ expectations.
“We don’t put brakes into a race car so we
can go slowly,” Kuperholz says.
“We put brakes into a race car so we can go
really quickly, but safely. Most organisations
have been using AI for a long time, even
if it’s simply through something like a
recommendation engine or a third-party piece
of marketing software.
“I think now it’s just very obvious that AI
is hitting the mainstream, and we will see
things like generative pre-trained transformers
being integrated into things like the Microsoft
Office Suite and the Google Suite, so they’ll be
landing on every desktop soon,” Kuperholz says.
AI AND ACCOUNTABILITY
Technology should never be unrestrained, but
who is ultimately accountable for generative
AI tools? Longstaff says it is a question yet to
be resolved.
