There is a growing fear that artificial intelligence could pose a threat not just to jobs, but also to factual accuracy, reputations, and even the very existence of humanity, which calls for striking a balance between safety and innovation in regulating AI.
Should we get
rid of all jobs, even the ones that make us happy? Should we make minds that
aren't human, which might outnumber, better, and replace us? Should we risk
letting our society get out of our hands? In an open letter sent out last month
by the NGO Future of Life Institute, these questions were asked. It asked that
the most advanced types of artificial intelligence (AI) be put on
pause for six months. Elon Musk and other big names in tech signed the
letter, and it is the best example yet of how fast growth in artificial
intelligence has made people worry about how dangerous it could be.
In
particular, new large language models (LLMS), like the ones that run
chatGPT, a chatbot made by a startup called OpenAI, have surprised even their
creators with skills they didn't expect. These emergent skills include
everything from being able to solve logic puzzles and write computer code to
being able to tell what a movie is about from an emoji story summary.
These models
could change how people interact with computers and machines, with
information, and even with themselves. AI supporters say that it could help
solve big problems by making new drugs, creating new materials to help fight
climate change, or figuring out how to make fusion power work. Others think
that the fact that AIS can do things its makers don't fully understand risks
bringing to life the science-fiction disaster scenario of a machine that
outsmarts its creator, which usually ends in death.
It is
difficult to balance the opportunities and hazards because of the seething
mixture of enthusiasm and anxiety that is currently present. However, there are
things that can be picked up from other industries as well as previous
transitions in technological paradigms. So, what exactly has evolved to make
artificial intelligence so much more capable? How frightened should you really
be? And what actions should be taken by governments?
Numerous
published pieces of evidence have investigated the inner workings of LLMS as
well as their plans for the future. When the initial wave of sophisticated AI
systems arrived a decade ago, they were dependent on data for training that had
been meticulously labeled. They might learn to do things like recognize photos
or transcribe speech after being shown a sufficient number of instances that
had labels attached to them. Modern systems do not need to be pre-labeled in
order to be trained, and as a result, they are able to utilize far larger data
sets that are acquired from internet sources. In practice, LLMS can be educated
on the entirety of the internet, which explains their powers, both positive and
negative.
The
release of chatGPT in November brought to the attention of a more general
audience the possibilities that were previously hidden. Within a week, one
million individuals had used it, and within two months, 100 million people had
used it. It wasn't long before it was being put to use in the
generation of school essays and wedding speeches. The success of chatGPT
and Microsoft's decision to integrate it into Bing, the company's search
engine, inspired other companies to develop and distribute chatbots of their
own.
The outcomes
of several of these were really peculiar. For instance, Bing Chat advised a
journalist that he should separate from his wife and start a new life
somewhere. A law professor has leveled the accusation of slander against
ChatGPT. LLMS generate answers that have the appearance of being factual but
frequently include inaccuracies or blatant fabrications of the facts. Despite
this, technology companies including as Microsoft, Google, and others have
begun incorporating LLMS into their products in order to assist customers in
the process of creating documents and carrying out other tasks.
The recent
surge in the power and visibility of artificial intelligence systems, as well
as the rising awareness of their capabilities and flaws, have stoked fears that
the technology is now progressing at such a rapid pace that it will be
impossible to govern in a secure manner. As a result, there has been a call for
a pause, and there is growing fear that artificial intelligence could pose a
threat not just to jobs, but also to factual accuracy and reputations, and even
to the very existence of humanity.
Regulating
AI: Striking a Balance Between Safety and Innovation
The worry
that robots will take over jobs goes back hundreds of years. But so far, new
technologies have created more jobs than they have taken away. Some jobs can be
done by machines, but not others. This means that there is more demand for
people who can do the jobs that machines can't do. Could it be different this
time? Even though there have been no signs of a quick change in the job market
so far, it is still possible. Before, technology tended to take over jobs that
didn't require much skill, but LLMS can do some white-collar jobs, like
summarizing papers and writing code.
There has
been a lot of talk about how much AI poses a grave risk. Experts don't agree.
In a poll of AI researchers done in 2022, 48% said that there was at least a
10% chance that the effects of AI would be extremely bad (like the end of
humanity). But 25% of researchers said there was no risk, and the median
researcher said the risk was 5%. The worst case scenario is that a very
smart AI does a lot of damage by making poisons or bugs or by getting people to
do terrorist acts. It doesn't have to be bad, but experts worry that future AIs
might have goals that are different from those of the people who made them.
Such
possibilities shouldn't be ruled out. But they all require a lot of guessing
and a big jump from what we know now. Many people think that in the future, AIs
will have unrestricted access to energy, money, and computer power, which are
real limits today and could be taken away from an AI that goes bad. Also, when
compared to other analysts, experts tend to exaggerate the risks in their own
area. (And Mr. Musk, who is starting his own AI company, has a reason to want
his competitors to fail.) Heavy rules or even a pause seem like an overreaction
right now. A pause would also be impossible to enforce anyway.
Regulation is
important, but not because it will save humanity. Concerns about bias, privacy,
and intellectual property rights are real when it comes to AI systems that are
already in use. As technology gets better, it might become clear that there are
other problems. The key is to weigh the benefits of AI against the risks and be
ready to change.
So far, three
different ways have been taken by states. On one end of the spectrum is
Britain, which wants to use a light-touch method that doesn't add any new
rules or regulatory bodies but does make sure that AI systems follow the rules
that are already in place. The goal is to get more people to invest and make
Britain a superpower in AI. The United States has taken a similar method, but
the Biden administration is now asking the public what a set of rules might
look like.
The EU is
getting stricter. Its suggested law puts different uses of AI into categories
based on how risky they are. As the risk goes up, from, say, recommending music
to self-driving cars, stricter monitoring and disclosure are needed. Some uses
of AI, like subliminal ads and biometrics that can be done from far away, are
outlawed. Companies that break the rules will have to pay a fine. Some critics
say that these rules are too restrictive.
But some
people say we need to be even tougher. Governments should treat AI like drugs,
with a dedicated regulator, strict testing, and pre-approval before it can be
used by the public. China is doing some of this by making companies register
their AI goods and go through a security review before putting them on the
market. But in China, politics may be a bigger reason than safety. For
instance, one of the most important requirements in China is that Ais'
work represent the core value of socialism.
How should
our society and governments react to this new trend? It's unclear that a
light touch will be enough. If AI is as important as cars, planes, and
medicines—and there are good reasons to think it is—then it will need new
rules, just like they did. So, the EU's model is the one that comes closest,
even though its classification system is too complicated and a method based on
principles would be more flexible. Requiring inspections and requiring
disclosure about how systems are taught, how they work, and how they are
monitored would be like rules in other industries.
This could
make it possible, if needed, to make the rules stricter over time. Then, a
dedicated regulator might seem like a good idea, as might international
treaties like the ones that rule nuclear weapons, if there is good evidence of
an existential risk. To keep an eye on this risk, governments could set up an
organization like CERN (in French Conseil Européen pour la Recherche
Nucléaire), a particle physics lab, that could also study AI safety and
ethics—areas where companies don't have as many reasons to spend as society
would like.
This strong
technology brings new risks, but it also gives us a lot of amazing chances. To
balance the two, the world have to be careful. Taking things slowly now
can lay the groundwork for more rules to be added in the future. But now is the
time to start building these foundations.
Notes
Agomuoh, F. (2023, February 13). Check Your
Inbox — Microsoft Just Sent Out the First Wave of ChatGPT Bing Invites. Retrieved
from Digital Trends:
https://www.digitaltrends.com/computing/bing-users-will-be-able-to-test-out-the-integrated-chatgpt/
CERN. (2023). About CERN. Retrieved from
https://home.cern/about
Future of Life Institute. (2023, March 22). Pause
Giant AI Experiments: An Open Letter. Retrieved from
https://futureoflife.org/open-letter/pause-giant-ai-experiments/
Grace, K. (2022, August 4). What Do ML
Researchers Think About AI in 2022? Retrieved from AI Impacts:
https://aiimpacts.org/what-do-ml-researchers-think-about-ai-in-2022/
Hu, K. (2023, February 2). ChatGPT Sets Record
for Fastest-Growing User Base - Analyst Note. Retrieved from Reuters:
https://www.reuters.com/technology/chatgpt-sets-record-fastest-growing-user-base-analyst-note-2023-02-01/
Huang, R. (2023, April 11). China Moves to
Censor AI. Retrieved from The Wall Street Journal:
https://www.wsj.com/articles/china-lays-out-strict-rules-for-chatgpt-like-ai-tools-32f70c89
Milmo, D. (2023, February 2). ChatGPT Reaches
100 Million Users Two Months After Launch. Retrieved from The Guardian:
https://www.theguardian.com/technology/2023/feb/02/chatgpt-100-million-users-open-ai-fastest-growing-app
Noorden, R. V. (2022). How Language-Generation
AIs Could Transform Science. Nature, 605(7808). doi:10.1038/d41586-022-01191-3.
PMID: 35484343.
O'Brien, M. (2023, March 29). Musk, Scientists
Call for Halt to AI Race Sparked by ChatGPT. Retrieved from AP News:
https://apnews.com/article/artificial-intelligence-chatgpt-risks-petition-elon-musk-steve-wozniak-534f0298d6304687ed080a5119a69962
The Economist. (2023, April 20). Technology
and Society: How to Worry Wisely About Artificial Intelligence. Retrieved
from
https://www.economist.com/leaders/2023/04/20/how-to-worry-wisely-about-artificial-intelligence
No comments:
Post a Comment