First, I enjoy how the speaker appeals to the human. In convention-style talks often there is this sort of ubermensch mentality. As in, 'How do we progress to the point where we are greater than what we already are?'. Here, she is saying that the human is just fine. We feel and experience incredible things. Rather than creating an AI that does things above the standards that are human, let us give the greatest gift to it, and that is our feelings and experiences. The gift of feeling human.
Next, we have quite a ways to go, but not really. For the AI to see a photo of a house burning down and respond with, 'This is an amazing view!', shows a complete lack of empathy. Far from being human. BUT! It does recognize the power and gravity of a raging fire. If the main point of the photo was a bonfire on the beach rather than a house, the AI would be correct.
To make sure we give the gift of humanity to our AI, we must first be honest with ourselves. We cannot share just the good. we must also share the mundane, and more importantly, the bad. Blemishes. Not just of humanity, but of the world. A house burning down shows no fault of humans. Its an unjust event that happens in the world. Here, the AI should show not amazement of the view, or even anger towards the unjustness, but rather the more complex emotion of compassion towards the grieving.
The first color images were calibrated against a white womans skin, which means that skin that is not white will show differently. The speaker uses the term blindspot for this phenomenon, a term I really like, because the job of a photo is to create the most accurate representation of a space in time. If it is not giving the proper tools to create the images, as in a full database of faces and images to draw from, the photo will do the job of recreating as best that it can, but will pull up short. Just as the AI will do the job of responding to the image of a fire, but will fail to fully empathize.
"AI can end mankind."
But only if we let it. The speaker mentions open-source tools that we all have access, too. Let's contribute! And first, let's define what AI is right now and what it CAN become. We still hold the cards when building AI. We have built some talented AI that is currently contributing to our lives quite positively, and some talented AI that is not so positive. In that global discussion about the present state of AI, lets be honest about both sides, learn the lessons, and implement them. This is how we leverage AI in the future, by making sure we learn from our past. We choose what the AI of the future looks like.
Unfortunately, it doesn't look like the current President has weighed in on the matter, but this shows that it has been on the mind of the White House. I believe that this should be a part of future gear-ups because, while it does go fairly in-depth about the particulars, it makes a nice, digestable recommendation list on page 40. Each of these could be examined and discussed.