The big news this week is definitely, Geoffrey Hinton leaving Google plus his indication that he is not sure if AI can be used for the good of humanity anymore. I wish him all the best and I hope he will be invited to sit in committees that regulate AI.
This is my view. Neural network (NN) going rogue has been around for almost a decade if I remember correctly. In the earlier years, we have NN vulnerable from adversarial attack. Don’t remember? The example is a "STOP" traffic sign that is recognizable by the NN. The same NN can be made to identify something else after making changes to certain pixels of the same "STOP" sign. There are examples where NN learned the wrong features such as the light background for tank photos or the snow in the background for wolves vs huskies recognition.
Subsequently, we have seen many other examples of data being abused such as Cambridge Analytica, or Google Photos, or Amazon Hiring. The abuse has been going around since the start. I just want to stress here, as long as you are using NN, you cannot run away from the downside that comes from the lack of transparency. All these has been around for almost a decade already.
Although there are these abuses, but we also gained many tools as well. For instance, your assistant found in the smartphone, or grandmasters using AI to improve their own game.
The previous two paragraphs is similar for any new tools that comes into humanity. Why is that so? Because tools can be used for the good of mankind or to harm mankind. There is no running away from it, all tools can be and will be abused. Let us not pretend that abuse will never happen.
Thus there is no need to be alarmed, imo. I just want to stress again that it is time to START having more and better regulations on how AI can be used and developed, and for businesses to start having Artificial Intelligence and Data Science representation in their board to reduce the potential abuse from data.
TLDR: There is no cause for alarm, really.
1) We need to start thinking about regulations, with both technical and non-technical sitting in these committees.
2) We need to have AI and DS representation in the Board. (Post)
3) We need a library of use cases, especially the abuses, which we do have. The database is indicated in my blog posts on AI Index recently. (Post 1) (Post 2)
4) We should look start looking at different ways to build AGI instead of building it through solely through the Neural Network architecture its lack of transparency cannot be avoided. (Post 1) (Post 2)
<Note> Geoffrey Hinton, although he should be given credit for continuing to pursue neural network as his primary research, plus training up many well-known researchers such as Ilya Sukstever and Russ Salakhutdinov but his last breakthrough to the development was Capsule Network which was around 2017. Its applications is not mentioned much though so his contribution in recent years did not made an impression on the public yet.
What are your thoughts? Please share them with me on LinkedIn. :)
Consider supporting my work by buying me a "coffee" here. :) Please feel free to link up on LinkedIn or Twitter (@PSkoo). I just started my YouTube channel, do consider subscribing to it to give support! Do consider signing up for my newsletter too.