Testing the Saftey of Smart Cars

Ivan Ang

Greetings from the Australian National University, where Ivan Ang is reporting his research on "User-centric Assessment of Robotic Systems via POMDP Planning of Adversaries". What got my attention was applying this to testing car autonomous driving. Recently I purchased a basic compact car. Despite the low cost, it came with a camera and radar for autonomous breaking, adaptive cruse control and lane keeping. I am trusting my life, and those of other road users, to these systems, so how well are they tested? Ivan pointed out that autonomous breaking was tested by pulling a dummy out in the path of the car on a string. The dummy's arms and legs are fixed, as its speed and direction. The developers of the system can optimize it for this test, but it then might do as well with real people. 

Calibrating AI For Saftey

Greetings from the Australian National University, where I attended the weekly AI, ML and Friends seminar. Jiawei Liu was speaking on "Uncertainty Calibration for Deep Neural Networks". I didn't understand the equations, but the idea, as I understand it, is to have a measure of how well AI is doing.

What got my attention was an example of recognising a speedboat. From the original photo it was clear to a human what this was. But the black and white outline, as used by some algorithms, looked to me like a USV (Uncrewed  Surface Vessel). These are being used, right now, in the Black and Read Seas, laden with explosives, to attack ships. The crews of warships have to maintain a lookout 24 hours a day, for weeks, trying to spot drone attacks. The USVs are modified speedboats so it is difficult to tell them from fishing boats. This type of AI analysis can help ensure mistakes are not made which could result in the loss of a ship, or the sinking of an innocent fishing boat.

Machine Learning, Games, Economic Systems and The End of the World

Greetings from the Australian National University where  Dr. Yun Kun Cheung is speaking on "Machine Learning in Games and Economic Systems" at the weekly AI ML & Friends seminar. One question raised was if algorithmic trading can cause stock markets to collapse, as well as deliberate attacks which incorporate algorithmic learning.

This is my second AI seminar of the day discussing the adverse effects of AI. The first was Arvind Narayanan & Sayash Kapoor (Princeton University) on "AI and Existential Risk". At question time I asked if the use of AI might be an acceptable risk, where it reduces the risk of nuclear war. As an example, the Australian Strategic Policy Institute released an article discussing the use of Australian nuclear powered submarines to defend Taiwan, and attack the Chinese mainland ("Punishing the Dragon—it’s not about Tomahawk missiles from SSNs", Malcolm Davis, 20 Mar 2024). Such a strategy risks escalating into nuclear war. As an alternative, I suggested small AI enabled drones, with limited offensive capability, could be used in place of nuclear submarines. As well as being more effective for defending, they have less capability for strategic offensive use.

Wide Angle View of Hybrid Classroom

Screenshot from AI, ML & Friends Seminar, 

Greetings from the weekly AI, ML and Friends Seminar, at the ANU School of Computing. Last week I was in the room, but this week I am zooming in. As well as the slides, the Zoom session features a wide angle view of the room, from the back. At first I wondered why, as you can't clearly see the speaker, or read what is on the board. But this turns out to be useful for getting an idea of what is going on. At the start of an event you can look at the image and see people are still milling around. 

Digital Legislation

Greetings from the Australasian National University where Dr. Guido Governatori is speaking on "Digital Legislation" at a AI, ML and Friends Seminar. The idea is to have legislation which computers can read and interpret as computer code. Dr. Governatori commented he was not talking about Robodebt, which failed due to a data problem, with social services not having tax data. He said he would talk about smart contracts, which are neither smart, or contracts. The claim is that this will reduce the burden on business, as an automated system can work though all the legal obligations. However, I suggest this might increase the burden, as it would allow much more, and more intricate legislation to be written. Also not only black letter law, issued by the legislature will be need to be encoded, but also case law from courts.