Tuesday, May 1, 2018

The Promise of Brain-Machine Interfaces: Recap of March's The Future Now: NEEDs Seminar

Image courtesy of Wikimedia Commons.
By Nathan Ahlgrim

If we want to – to paraphrase the classic Six Million Dollar Man – rebuild people, rebuild them to be better, stronger, faster, we need more than fancy motors and titanium bones. Robot muscles cannot help a paralyzed person stand, and robot voices cannot restore communication to the voiceless, without some way for the person to control them. Methods of control need not be cutting-edge. The late Dr. Stephen Hawking’s instantly recognizable voice synthesizer was controlled by a single cheek movement, which seems shockingly analog in today’s world. Brain-machine interfaces (BMIs) are the emerging technology that promise to bypass all external input and allow robotic devices to communicate directly with the brain. Dr. Chethan Pandarinath, assistant professor of biomedical engineering at Georgia Tech and Emory University, discussed the good and bad of this technology in March’s The Future Now NEEDs seminar: "To Be Implanted and Wireless". He shared his experience and perspective, agreeing that these invasive technologies hold incredible promise. Keeping that promise both realistic and equitable, though, is an ongoing challenge.

BMIs are currently designed as assistive technologies. They can take many forms: a cochlear implant, a cursor on a screen, a robotic arm, or even a complete exoskeleton. All serve the same general purpose: to restore a person’s ability to connect and communicate with the world. The most common patients are those with some form of paralysis. Given the potential to restore movement or speech to people, many see the development of BMIs as a moral imperative. However, agreeing that BMI research is a worthwhile and necessary endeavor cannot will these devices into being. There is a good reason why controlling a robot arm with your brain feels like something out of science fiction – it is incredibly difficult to do.

An example of an intracortical array.
Image courtesy of Wikimedia Commons.
Reliable BMIs depend on first being able to record brain activity. Scientists have been able to do this for decades at great precision, but the unfortunate trade-off is that the level of precision tracks directly with the level of invasiveness. As Dr. Pandarinath described, scalp electroencephalograms (EEGs) require no surgery at all, but analyzing the resulting data is like standing outside of a football stadium. You may hear the roar of the crowd, but you need to get in the stands before you can pick up individual conversations. For scientists, that means you need to open up the skull and place arrays of wires (known as intracortical microelectrodes) into the brain itself in order to eavesdrop on the brain’s conversations.

Display of the BrainGate system.
Image courtesy of Wikimedia Commons.
Figuring out what those brain conversations mean is the hard part. All our billions of neurons firing at once produce gigabytes of data, and the challenge of making sense of that data is what draws engineers and computer scientists towards neuroscience. Dr. Pandarinath is one of these people, a self-described “engineer that managed to run into the brain one day and thought it was pretty cool.” Approaching the problem as an engineer, he and many others have developed a host of technologies around the BrainGate system. Their tagline says it all: “Turning thought into action.” Targeting the motor cortex of the brain, which controls voluntary movements in healthy individuals, BrainGate technology allows paralyzed people to control robotics just by thinking about them (Pandarinath et al., 2017). Perhaps most shocking of all, learning to control the device is like learning to walk. At first it’s a struggle (there’s a reason we label toddlers as such), but adults do not consider walking a skill. As one patient described, “it was hard work getting [it to work]. I struggled greatly to [move the arm] up and down at the beginning, now up and down is so easy I don’t even think about it.” In effect, BrainGate lets patients control a robot as an extension of their own body. No mental gymnastics needed.

Is the ease of use a good thing? Once patients can “automatically” control BMIs, are they at fault for any harm caused by the machine? Dr. Karen Rommelfanger raised one possible scenario: following an argument between the patient and researcher, the patient’s robotic hand crushes the researcher’s hand during testing. Who is at fault? Did the patient misuse the technology, or did the researcher cause her own injury by creating a faulty system?

One possibility is to have a universal limit to the strength and ability of all BMIs. Even though we can create machines that rip cars apart like tissue paper, maybe we should never build a robotic arm to have more grip strength than that of a child. Such a solution prevents the person (or BMI) from doing any physical harm, but it then fails the primary goal of BMIs: to restore patients’ abilities. A universal set-point on what these abilities should be is problematic because, for better or worse, there is no singular ‘human ability.’

By the end of the seminar, the conversation landed on where to draw the line between restoration and enhancement. Of course, this debate is not new to BMIs. Everything from sports supplementation to psychostimulants like Adderall are subject to the same debate: who deserves to receive these treatments, and how much is too much? Researchers do not even need to design superhuman BMIs (although it is certainly possible) to join the conversation. The arm strength of an editorial intern is a far cry from Game of Thrones’ Hafþór Björnsson, but we are both decidedly human. If I became paralyzed, must I be restricted to my previous strength? I could always argue that I was just going to start a strongman program before I became paralyzed, and therefore I deserve a robotic arm to match.

Could and should BMIs make everyone
as strong as humanly possible?
Image courtesy of Wikimedia Commons.
The premise that researchers will be in charge of setting a limit (if any) may be inherently flawed, given that machine learning is starting to drive BMI research. Algorithms succeed by optimizing solutions, which in the case of BMIs would mean the most efficient, the most precise, and perhaps the strongest BMI possible. Normal humans are hardly the optimal physical form, so it is hard to imagine a sophisticated algorithm being complacent at returning me to my previous strength.

To many, “supplementing” people with artificial intelligence (AI)-guided BMIs is a good thing, and perhaps even necessary. Elon Musk, famous for his dire warnings on the impending AI threat, posits that coupling AI with humans via BMIs is the best protection our species has against it. By making ourselves more than human, we will at least have a fighting chance against the AIs we design with the express goal of being better than human.

In the end, BMIs do offer great promise. No, a paraplegic will not be able to walk normally in the next year using a BMI. Anyone who promises that is peddling in false hope and unrealistic expectations. But BMIs, like all other technologies, never stop improving. Questions about limits to and access to these incredible tools will only become more pressing as the technology improves. Who gets to set the limit? Who will act as gatekeeper? The patient or the manufacturer? Dr. Pandarinath does not think BMIs are different than any other cutting-edge product: “by default, it’ll be the wallet.” And adjusting for inflation, it will now take thirty-five million dollars to build the Six Million Dollar Man.

References

Pandarinath C, Nuyujukian P, Blabe CH, Sorice BL, Saab J, Willett FR, Hochberg LR, Shenoy KV, Henderson JM (2017) High performance communication by people with paralysis using an intracortical brain-computer interface. eLife 6:e18554.

Want to cite this post?

Ahlgrim, N. (2018). The Promise of Brain-Machine Interfaces: Recap of March's The Future Now: NEEDs Seminar. The Neuroethics Blog. Retrieved on , from http://www.theneuroethicsblog.com/2018/05/the-promise-of-brain-machine-interfaces.html

No comments: