Free beta access. 50 free AI tokens each week. Optional top-ups available. See what is included
Opening Word document…

Haenel 18

June, Fabio Haenel currently works at the Global and European Studies Institute, University of Leipzig. Fabio does research in Information Technology and Politics, Human Rights and Foreign Policy. “The Prospects and Dangers of Artificial Intelligence on International Security: The Case of a Sino-American Arms Race.”, https://www.researchgate.net/publication/325542003_The_Prospects_and_Dangers_of_Artificial_Intelligence_on_International_Security_The_Case_of_a_Sino-American_Arms_Race – ECM

Multiple arguments for the rapid development of Artificial Intelligence outpace any skepticism towards the fledgling technology, their pressing nature however showcases the necessity to do so safely. First, delaying the advancement of a technology that could help to solve contemporary and future social and scientific problems, e.g. by increasing the efficiency of energy consumption or outright help to invent entirely new methods and materials to produce sustainable energy, or by reducing traffic through automatic and interconnected private and public transportation systems, carries a high opportunity cost to delay these achievements to occur later, or never, rather than sooner.43 This becomes even more apparent if we consider the AI-assisted scientific breakthroughs such as in medicine where machines already outperform certified and experienced doctors in detecting selected diseases accurately to the benefits o the patients.44 Second, stalling investments in research and development of AI by individual powerful actors will only give advantage to a competing state or corporation which is willing to fund and pioneer the technology. Reasons to delay include the monetary costs o innovating a field dependent on highly-skilled labor and benefiting most rom an academic and industry in restructure that is locally accessible, or the lack o conviction of its use in the near future among term- or profit-constrained government and business leadership.

These contradictions reveal a striking paradox of the technology’s geopolitical quality: Similar to the development and manipulation of nuclear power in the 1940s, succeeding first in realizing powerful AI capable of prioritizing the economic or strategic interests of its developers will reset the political spheres of interests globally, and thus now has captured the attention of regional and global political and economic powers. However, newly developed software can also be much more easily misused by rogue actors than nuclear weapons because of its decentralized, inexpensive accessibility, and any further delay in developing AI robustly could invite “bad actors” such as states, terrorist groups, or corporations to exploit it unilaterally.

Research on AI safety thus is necessary to “help maximize the societal benefit of AI,” both to ensure that the technology is utilized to ease and enrich human lives while it caters to the political and economic interests driving its adoption forward notwithstanding the paramount importance to guarantee that “AI systems must do what [humans] want them to do.”45 Among experts it is taken or granted that its industrial development will be progressed by capital investments ignorant of policy adjustments, thus theoretical and practical research is needed to better understand “the nature, risks, and overall outcomes” and safeguard specific and general AI applications.46 These concerns are nourished by the logical assumption, borrowing from earlier explanations why not to anthropomorphize machines, that enabling a machine to automatically re-program itself would lead to the departure of its original programming within the rules set or its auto-improvement by the original programmers. Russel argues that these rules mark the critical juncture between developing AI safely or ailing to do so because of the virtually infinite density of the rules’ consequences on the machines’ future actions.

Bostrom illustrates this with an anecdote of tasking an intelligent machine with generating as many paperclips as possible to ease a supply shortage.48 However, the wording includes not a finite number of clips or how it should and should not achieve its goal. Thus, besides collecting them individually, trading them or money generated from profits earned in other capacities, the AI will eventually seek to control the manufacturing of paperclips itself .49 To maximize its utility function the goal-seeking AI soon will put such factories across all available space on Earth, effectively eradicating any habitable space and with it all life on the planet. The machine carried out its goals within the rules sets by its human developers without knowing how many clips are enough, or whether sustaining life is more important than creating paperclips.50 Researching how to safely align such most critical human interests with binary operations is integral to the robust development of Artificial Intelligence, rendering this research “the essential task of our age”.

Safety advocates argue that “whether the risk is 1 per cent, 80 per cent, or anywhere in between” it seems irresponsible not to take any precautions.56

Besides, predicting the impossibility of any technology that has yet to be developed practically appears to be flawed: Ernest Rutherford, a respected physicist o his time, proclaimed at a dinner reception in September 1933 that “anyone who predicted energy could be derived rom the trans ormation o an atom ‘was talking moonshine.’ The very next day […] Leo Szilárd worked out conceptually how it could be done.”57 Szilárd’s discovery eventually led to the manipulation of nuclear power or civil and military purposes. In other words, a technology deemed inconceivable turned out to be the base of the most constructive and destructive force ever commanded by humans in the very next moment.

Sign in to recut, update, or save this card.