Book review: Genesis: Artificial Intelligence, Hope, and the Human
Spirit, by Henry A. Kissinger, Eric Schmidt, and Craig Mundie.
Genesis lends a bit of authority to concerns about AI.
It is a frustrating book. It took more effort for me read than it should
have taken. The difficulty stems not from complex subject matter
(although the topics are complex), but from a peculiarly alien writing
style that transcends mere linguistic differences - though Kissinger's
German intellectual heritage may play a role.
The book's opening meanders through historical vignettes whose
relevance remains opaque, testing my patience before finally addressing
AI.
Risks
When the book gets around to discussing how AI will affect our future,
it's mostly correct about AI being a big deal, with occasionally
appropriate hints about why there are big risks. But it's frustratingly
abstract and vague. Some examples:
we might become extinct.
Would networking intelligences make their processes more opaque than
the processes of lone intelligence? ... would we be able to assess
them on a spectrum of good to evil? Or would they operate on an
informational basis - extracted at superhuman speed ... - that would
confound our ability to judge their behavior? Would that lead us
further into a cycle of passivity?
Today, in the years, months, weeks, and days leading up to the arrival
of the first superintelligence, a security dilemma of existential
nature awaits.
I see hints in that quote that they think the threshold of
superintelligence will be well enough defined that it can be attributed
to a specific day. I find that suspicious.
Genesis compares our preparedness for AI to the preparedness of Aztecs
for the arrival of conquistadors.
One area where the book briefly feels clear and novel is when it
discusses the future of war, notably observing that humans may become
less targeted simply because they'll be irrelevant to military
outcomes.
The book provides only weak hints as to what considerations are
important. It often feels like there's a missing
mood -
e.g. it's hard to tell whether the authors think human extinction would
be a bigger deal than the end of democracy.
Present Day AI
The weakest parts of the book attempt to describe current AI. Too many
of those claims look like claims that were discredited several years
ago. It was published a year after Kissinger's death, so likely some of
the problem is a long delay between when he wrote those parts and
publication.
But there will be phases in the evolution of AI when mechanical
intelligence may feel eerily similar to the intelligence of the
animals.
I'd say that "prediction" was plausibly true of the best AIs for a
brief time around 2021 or 2022. Now AIs seem more like human children.
Lately, AI researchers have devoted serious attention to the project
of giving machines "groundedness" - a reliable relationship between
the machine's representations and reality
This was true in 2022, but it has been increasingly treated as a solved
problem since then.
Other Thoughts
Will we become more like them, or will they become more like us? ...
Answering it remains our first and most necessary task.
The authors express cautious optimism about brain-computer interfaces
facilitating human-AI symbiosis. That suggests either an overestimation
of neural interface potential or an underestimation of AI's rapid
advancement.
Under this definition, can AI itself possess dignity? Likely not - for
AIs are not born, do not die, feel neither insecurity nor fear, and do
not have natural inclinations or individuality such that conceptions
of evil or good could be considered "theirs". ... they should be
treated, philosophically, like literary characters.
This feels like a confused mix of half-assed morality and limited
understanding of where AI is headed.
Genesis refers to Nick Bostrom and Eliezer Yudkowsky without criticizing
them. Combined with Kissinger's reputation, that will cause some
political and military leaders to take the risks of AI more seriously.
That makes the book somewhat important.
People should read this book if they respect Kissinger's forecasts much
more than they respect the forecasts of people connected with tech
companies.
Book review: Genesis: Artificial Intelligence, Hope, and the Human Spirit, by Henry A. Kissinger, Eric Schmidt, and Craig Mundie.
Genesis lends a bit of authority to concerns about AI.
It is a frustrating book. It took more effort for me read than it should have taken. The difficulty stems not from complex subject matter (although the topics are complex), but from a peculiarly alien writing style that transcends mere linguistic differences - though Kissinger's German intellectual heritage may play a role.
The book's opening meanders through historical vignettes whose relevance remains opaque, testing my patience before finally addressing AI.
Risks
When the book gets around to discussing how AI will affect our future, it's mostly correct about AI being a big deal, with occasionally appropriate hints about why there are big risks. But it's frustratingly abstract and vague. Some examples:
I see hints in that quote that they think the threshold of superintelligence will be well enough defined that it can be attributed to a specific day. I find that suspicious.
Genesis compares our preparedness for AI to the preparedness of Aztecs for the arrival of conquistadors.
One area where the book briefly feels clear and novel is when it discusses the future of war, notably observing that humans may become less targeted simply because they'll be irrelevant to military outcomes.
The book provides only weak hints as to what considerations are important. It often feels like there's a missing mood - e.g. it's hard to tell whether the authors think human extinction would be a bigger deal than the end of democracy.
Present Day AI
The weakest parts of the book attempt to describe current AI. Too many of those claims look like claims that were discredited several years ago. It was published a year after Kissinger's death, so likely some of the problem is a long delay between when he wrote those parts and publication.
I'd say that "prediction" was plausibly true of the best AIs for a brief time around 2021 or 2022. Now AIs seem more like human children.
This was true in 2022, but it has been increasingly treated as a solved problem since then.
Other Thoughts
The authors express cautious optimism about brain-computer interfaces facilitating human-AI symbiosis. That suggests either an overestimation of neural interface potential or an underestimation of AI's rapid advancement.
This feels like a confused mix of half-assed morality and limited understanding of where AI is headed.
Genesis refers to Nick Bostrom and Eliezer Yudkowsky without criticizing them. Combined with Kissinger's reputation, that will cause some political and military leaders to take the risks of AI more seriously. That makes the book somewhat important.
People should read this book if they respect Kissinger's forecasts much more than they respect the forecasts of people connected with tech companies.