AKA Mimetic Supremacy
IBM's AI debating system able to compete with expert human debaters
Mar 2021, phys.org
The IBM system, known simply as Project Debater, scans the internet for such arguments and uses them in ways that it has learned are convincing.
Project Debater was asked to convince a panel of viewers that telemedicine was a good idea. Most of those on the panel found that the AI system did indeed change their stance on the topic—a possible indication that AI systems may one day soon play a role in human debates such as those that occur on social media sites.
via: Noam Slonim et al. An autonomous debating system, Nature (2021). DOI: 10.1038/s41586-021-03215-w
SEIHAI - The hierarchical AI that won the NeurIPS-2020 MineRL competition
Dec 2021, phys.org
The winning algorithm was called "sample-efficient hierarchical AI" or SEIHAI, and it combines machine learning, human observation, and more traditional computation tasks. The key is to break down the overall task into subtasks, and then finding examples of humans doing the subtasks, and from that, finding the best way to do each subtask. They significantly outperformed their competitors.
via Huawei Noah's Ark Lab at Tianjin University and Tsinghua University, winners of the Neural Information Processing Systems (NeurIPS) annual conference: Hangyu Mao et al, SEIHAI: A sample-efficient hierarchical AI for the MineRL competition. arXiv:2111.08857v1 [cs.LG], arxiv.org/abs/2111.08857
Researchers demonstrate first human use of high-bandwidth wireless brain-computer interface
Apr 2021, phys.org
All Brains Ahead
Participants with tetraplegia have demonstrated use of an intracortical wireless BCI with an external wireless transmitter. The system is capable of transmitting brain signals at single-neuron resolution and in full broadband fidelity without physically tethering the user to a decoding system. This wireless system is functionally equivalent to the wired systems that have been the gold standard in BCI performance for years,
via BrainGate and Brown University: John D Simeral et al. Home Use of a Percutaneous Wireless Intracortical Brain-Computer Interface by Individuals With Tetraplegia, IEEE Transactions on Biomedical Engineering (2021). DOI: 10.1109/TBME.2021.3069119
Engineers are designing an autonomous robot that can open doors, find nearest electric outlet
Nov 2021, phys.org
With a headline like that, why bother reading the article?
via University of Cincinnati: Yufeng Sun et al, Force-Vision Sensor Fusion Improves Learning-Based Approach for Self-Closing Door Pulling, IEEE Access (2021). DOI: 10.1109/ACCESS.2021.3118594
'Deepfaking the mind' could improve brain-computer interfaces for people with disabilities
Nov 2021, phys.org
When I tried to use the EPOC headset many years ago, I was faced with an existential quandry. You're supposed to put on the headset, and control a simple video game with your thoughts. But in order to prepare your headset, you have to teach it the first and most important lesson, which is the difference between having a thought and not having a thought.
Your brain isn't a quiet place, there's electric fields bouncing around all over the place, all the time, and in order to separate the signal from the noise, your headset has to known what is considered non-thought, "default state" activity. What does your brain sound like when it's not giving a command. Once we have that default baseline state, we can detect when you're giving an active, intentional command. Later on, we can start to differentiate commands, but to begin with, we just need to know what a command looks like, and what it doesn't.
To do this on the superscale, we would need to give the headset a whole lot of baseline data, which would require lots of different people to all wear these headsets all day.
Now, it looks like there's a way to make synthetic brainwave data, using generative adversarial networks (GANs), and then feeding it to the program. What used to take 20 minutes now takes 1 minute.
via University of Southern California: Shixian Wen et al, Rapid adaptation of brain–computer interfaces to new neuronal ensembles or participants via generative modelling, Nature Biomedical Engineering (2021). DOI: 10.1038/s41551-021-00811-z
Wireless network controls brain circuits remotely via the internet
Dec 2021, phys.org
Relax, nothing to worry about
A new study shows that researchers can remotely control the brain circuits of numerous animals simultaneously and independently through the internet.
The wireless ecosystem only requires a mini-computer that can be purchased for under $45, which connects to the internet and communicates with wireless multifunctional brain probes or other types of conventional laboratory equipment using IoT control modules.
"They can remotely perform large-scale neuroscience experiments in animals deployed in multiple countries," said one of the lead authors, Dr. Raza Qazi, a researcher with KAIST and the University of Colorado, Boulder.
via The Korea Advanced Institute of Science and Technology (KAIST), Washington University in St. Louis, and the University of Colorado, Boulder: Raza Qazi et al, Scalable and modular wireless-network infrastructure for large-scale behavioural neuroscience, Nature Biomedical Engineering (2021). DOI: 10.1038/s41551-021-00814-w
AI-generated faces found more trustworthy than real faces: Researchers warn of 'deep fakes'
Feb 2022, phys.org
Mimetic Supremacy Achieved
"Our evaluation of the photo realism of AI-synthesized faces indicates that synthesis engines have passed through the uncanny valley and are capable of creating faces that are indistinguishable—and more trustworthy—than real faces."
via Lancaster University: Sophie J. Nightingale et al, AI-synthesized faces are indistinguishable from real faces and more trustworthy, Proceedings of the National Academy of Sciences (2022). DOI: 10.1073/pnas.2120481119
DeepMind AI rivals average human competitive coder
Feb 2022, BBC News
Almost there
After simulating 10 contests, with more than 5,000 participants, AI system AlphaCode has ranked in the top 54% of competitors. There was still work to do to bring it up to the same level as top performing humans, DeepMind said.
