4 ms·

There is nothing fundamentally different between AI and “traditional algorithms” when it comes to ethical considerations. Sure, the current crop of AI systems p

by GuiA 8y ago

There is nothing fundamentally different between AI and “traditional algorithms” when it comes to ethical considerations. Sure, the current crop of AI systems perform unbelievably well on tasks that “traditional algorithms” suck at, but it is a change in degree, not in kind. In the overwhelming majority of pieces reflecting on AI ethics, you can replace “AI” with “algorithms”, “big data”, “computers”, “technology”, etc and not get a significant change in meaning. Many people have had their livelihood, rights, image, etc attacked in some way by technologically mediated processes that were deployed by organizations that didn’t think things through, or prioritized eg profits over other concerns, or had anti-humanist agendas. This phenomenon is certainly accelerated by the ever increasing ubiquity of computing systems, but it certainly predates them. AI does not bring anything radically different to the table in that regard, and if Google does not seem to care much about the ethical implications of their technology in general (see for instance the YouTube/children fiasco), why would they care about AI ethics specifically?

naasking 8y ago
I think you're wrong. The tooling around training ML is becoming progressively more accessible, so much so that within ten years you won't even need a programmer to train an AI for many tasks. Who knows where that will be in 20 years.
- fwn 8y ago
  But what you're saying, which seems to be that technology is getting more accessible over time, does not appear to me to be fundamentally different for other technological means, be it algorithms, ai or digitalization in general.
  - naasking 8y ago
    It is different. Algorithms require understanding what the algorithm does, how it operates. Lay people can't do this without significant training. AI/ML in principle simply requires example mappings of inputs and outputs. Lay people can do this.

by GuiA 8y ago

naasking 8y ago

I think you're wrong. The tooling around training ML is becoming progressively more accessible, so much so that within ten years you won't even need a programmer to train an AI for many tasks. Who knows where that will be in 20 years.

fwn 8y ago
But what you're saying, which seems to be that technology is getting more accessible over time, does not appear to me to be fundamentally different for other technological means, be it algorithms, ai or digitalization in general.
- naasking 8y ago
  It is different. Algorithms require understanding what the algorithm does, how it operates. Lay people can't do this without significant training. AI/ML in principle simply requires example mappings of inputs and outputs. Lay people can do this.