More on this book
Community
Kindle Notes & Highlights
Every society has, in its own way, inquired into the nature of reality: How can it be understood? Predicted? Shaped? Moderated?
The invisible governors of the moral world were inevitably cast in a similar mould of fiction and allegory.
The invention of the printing press made it possible to circulate materials and ideas directly to large groups of people in languages they understood rather than in the Latin of the scholarly classes, nullifying people’s historic reliance on the church
new ideas could be spread and amplified faster than they could be restricted.
When information is contextualized, it becomes knowledge. When knowledge compels convictions, it becomes wisdom.
But as it is being applied to more elements of our lives, it is altering the role that our minds have traditionally played
But machine-learning algorithms are a departure from the precision and predictability of classical algorithms, including those in calculations like long division.
which consist of steps for producing precise results, machine-learning algorithms consist of steps for improving upon imprecise results.
went from attempting to encode human-distilled insights into machines to delegating the learning process itself to the machines.
different goals and functions require different training techniques.
(As of this writing, AI has detected breast cancer earlier than human doctors by identifying subtle radiological indicators; it has detected retinopathy, one of the leading causes of blindness, by analyzing retinal photos; it has predicted hypoglycemia in diabetics by analyzing medical histories; and it has detected other heritable conditions by analyzing genetic codes.)
This process, training AIs on roughly matching — but untranslated — bodies of text, is the parallel corpora technique.
cannot reflect; the significance of its actions is up to humans to decide. Humans, therefore, must regulate and monitor the technology.
overestimates oxygen saturation in dark-skinned individuals.
society cannot mitigate what it does not foresee.
AI’s brittleness is a reflection of the shallowness of what it learns.
AI to be employed only after its creators demonstrate its reliability through testing processes.
Developing professional certification, compliance monitoring, and oversight programs for AI — and the auditing expertise their execution will require — will be a crucial societal project.
App developers often rush programs to market, correcting flaws in real time, while aerospace companies do the opposite: test their jets religiously before a single customer ever sets foot
Because an AI’s evolution halts after training, humans can assess its capacities without fear that it will develop unexpected, undesired behaviors after it completes its tests.
a self-driving car trained to stop at red lights cannot suddenly “decide” to start running them. This property makes comprehensive testing and certifications possible — engineers
And if the programmer does not put the capacity there, or explicitly forbids the action, the AI cannot do it. Second, AI is constrained by its objective
Training data, in turn, requires substantial computing infrastructure, making retraining AI prohibitively expensive, even if it is otherwise desirable to do so.
With data and computing requirements limiting the development of more advanced AI, devising training methods that use less data and less computer power is a critical frontier.
Language-translation AI stagnated for decades, then, through a confluence of techniques and computing power, advanced at a breakneck pace.
It is reasonable to expect that over time, AI will progress at least as fast as computing power has, yielding a millionfold increase in fifteen to twenty years.
Likely, development will yield AI “savants” — programs capable of dramatically exceeding human performance in specific areas, such as advanced scientific fields.
creators. But the anxieties underlying such common fantasies are mistaking the issue by assuming that AI’s culmination will be to act like individual humans.
To compound the challenge, once AI has been trained, it typically acts faster than the speed of human cognition.
To achieve greater convenience and accuracy, human developers have had to willingly forgo a measure of direct understanding.
People naturally gravitate toward existing gatherings, which leads to larger aggregations of users.
Rather than clear outcomes, however, we are more likely to arrive at a series of dilemmas with imperfect answers.
Small differences in the design of an AI’s objective function, training parameters, and definitions of falsehood could lead to society-altering differences in outcome.
To stay at the evolving forefront of technological development requires intellectual and financial capital beyond what most companies possess — and beyond what most governments are willing or able to provide.
In Beijing, Washington, and some European capitals, concern has been expressed (and articulated obliquely elsewhere) about the implications of conducting broad aspects of national economic and social life on network platforms facilitated by AI designed in other, potentially rival, countries. From
Yet Europe continues to face disadvantages for the initial scaling of new network platforms because of its need to serve many languages and national regulatory apparatuses
may elect either to chart a more independent path or assume a principal role within an international bloc of technologically compatible nations.
If a network platform is useful and successful, it comes to support broader commercial and industrial functions — and, in this capacity, it may become nationally indispensable.
This regulatory approach may create tension with a population’s expectation that it should be free to use whatever works best. In open societies, such prohibitions may also raise difficult and novel questions about the proper scope of government regulation.
Questions that appear urgent may be out of date by the time the relevant official participants have gathered to discuss them.
Dialogue between these sectors about core concerns and approaches is urgently needed — and
they should recognize that prevailing requires a definition of success that a society can sustain over time.
What degree of inferiority would remain meaningful in a crisis in which each side used its capabilities to the fullest?
When a crisis of no inherent global significance arose — the assassination of the heir to the Habsburg throne by a Serbian nationalist — the great powers of Europe followed these plans into a general conflict.
AI may permit simultaneous translation or the instantaneous relay of other critical information to personnel in crisis zones,
Although creating a sophisticated AI requires substantial computing power, proliferating or operating the AI generally does not.
no policy maker could know what would follow a “limited” use and whether it would remain limited.
This dynamic depended both on a state’s physical capacities and on an intangible quality: the potential aggressor’s state of mind and its opponent’s ability to shape
Viewed through the lens of deterrence, seeming weakness could have the same consequences as an actual deficiency;
a bluff taken seriously could prove a more useful deterrent than a bona fide ...
This highlight has been truncated due to consecutive passage length restrictions.
