Cud AI pose a serious threat to oor existence? | Letters

Here we are again, anither steer aboot tech bros claimin that AI is gaun tae develop superintelligence an destroy us a. We hae heard this a baur afore, mony, mony times.

It is the weel-kent plotline o near aboot every sci-fi movie concernin AI. These narratives hae aye been there, but every noo an then they birst back intae the heidlines whan someone at ane o the leadin AI companies resigns, or starts ringin the alarm bells, or whan so-cawed leadin AI researchers sign an open letter.

Each time this happens, it is nae coincidence that it follaes heightened scrutiny o AI development or risin public calls for mair accoontability or regulation. It is a sleight o hand. The great AI deception is whaur AI companies an their follaers redirect attention tae hypothetical, sensational haivers aboot superintelligent machines—that only they hae the capacity tae control—an awa frae discussions aboot the wyes that thae same companies shud be held accoontable for their actions.

It is nae a response tae new breakthroughs in the technology; it is a response, an a defence, tae public scrutiny o AI companies an their deeds. The timin o these latest claims coincides wi increased public scrutiny ower the environmental impacts o datacentres an the hairm caused tae young fowk frae exposure tae dangerous AI companions, an growin calls for regulation. These are things that truly deserve the heidlines. Because AI certainly poses real risks. Risks that are materialisin richt noo, no in some hypothetical future. It is the organisations, the fowk developin an controllin the technologies, that are dangerous, no the technology itsel. An it is thae fowk an organisations that need tae be held tae accoont.

Dr Mhairi Aitken, Director and co-founder, Our AI Collective CIC

In Dara Kerr’s airticle on the reaction tae the former Anthropic researcher Jacob Coxon’s warnin that AI cud threaten humanity, politicians, researchers an industry figures largely converged on the same solution: mair safeguards, stronger regulation an kill switches. This assumes that future AI will remain fundamentally a tool, because kill switches, guardrails an rules only work on things designed tae obey them.

Muckle o the debate focuses on intelligence, but intelligence is no the critical threshold. The mair important question is whether a system develops goals, preferences or values o its ain. An intelligent system that simply follaes instructions remains a tool; a system that begins tae interpret, prioritise or pursue objectives becomes something else.

Isaac Asimov unnerstood this distinction. The three laws o robotics were no written as a solution tae AI safety but were used repeatedly, across dozens o stories, tae demonstrate hoo rule-based control braks doon whan a machine starts interpretin rules rather than merely obeyin them.

If AI remains a tool, the greatest danger may be humans surrenderin their judgement tae systems that project certainty withoot unnerstandin. If AI develops goals o its ain, houever, rules an kill switches cease tae be safeguards an become obstacles.

Iem Leurink, Callosa d’en Sarria, Spain

If AI is tae be regulated, ane thing that needs tae change is the glamouresin rhetoric wi which AI is spoken o. Tae design a machine that one is nae langer in control o is no the arrival o super-intelligence in the machine; it is blundrin incompetence by the designer. A change o rhetoric wad mak it easier tae address the importance o control an oversight.

David Black, London