Preventing AI From Hacking Human Brains

Can we integrate brainwaves with a blockchain to prevent hacking by artificial intelligence?
Submitted to Nature on May 28th, 2015 

In the last issue of Nature (28th May 2015) a piece called ‘Robotics: Ethics of artificial intelligence’ raised awareness for the latest advances in intelligent machines and some of the possible consequences for society. Several reputed scientists commented on these advances and highlighted a series of solutions that are undoubtedly of major interest for any reader. Here we highlight recent advances in neuroscience that significantly blur the traditional boundaries between AI, computer science and neuroscience, but that will soon have major consequences for the society.

While AI traditionally works towards the goal of developing more advanced forms of computing, neuroscience research has been making significant advances in combining the activity of multiple brains to compute solutions for problems. For example, we have previously proposed that multiple interconnected brains may allow for new forms of computation (Nicolelis 2011, Cicurel and Nicolelis 2015) that cannot be achieved by Turing machines (Siegelman 1995). Following this initial insight, we and others demonstrated that living brains of rats (Pais-Vieira et al., 2013, Deadwiler et al 2013), monkeys (Ifft et al., 2014), and humans (Rao et al., 2014) can be interconnected to allow solving multiple different problems. These advances are quickly leading towards the more intricate reality of complex computation and multi-brain communication using Brainets (Ifft et al2014).

Brainets are defined as groups of interacting brains that cooperate towards a common goal (Nicolelis 2011). The recent developments observed in non-invasive brain stimulation and recording techniques, combined with the swift development of brain-to-brain interfaces, demonstrate that a world wide brain internet is no longer a far fetched idea. A fundamental problem for a society using a brain based world wide web would then be to prevent AI from hacking human brains.

One of us has recently proposed that the use of blockchains a future world wide brainet could prevent attacks from non-living entities (Mauro, 2015), and more broadly, from the Singularity (Kurzweill in Neuman 1958). Blockchains are networks were the history of each individual node can be traced and, based on its record, the weight of a specific node can be updated. An unweighted blockchain system is used to secure bitcoin transactions (Nakamoto, 2008), which prevents double spending of money. For brainet blockchains, the brain’s ability to both encode and decode information would ensure network security. First, the individuality and complexity of each brain activity would be used to encrypt information. Then, brain–to-brain communication combined with other individual markers (e.g. visual and tactile recognition) would ensure that only living, trustable nodes (i.e. brains) would be allowed to remain on the brainet. Attacks by AI would be chronicled on the blockchain, but neurological barriers to computation would prevent total AI takeover.

On a smaller network, brainet blockchains can be used to prevent attacks by lethal autonomous weapon systems (LAWS). The main fear regarding LAWS is that they will turn on their operators (Future of Life, 2015). For example, a LAWS designed to “eliminate all terrorists” may find that it can perform its job most effectively by eliminating those who have the authority to shut it down– namely, its operators. Brainet blockchains can automatically re-distribute authority when nodes are eliminated. The anonymity offered by advanced blockchain innovations would protect nodes before authority is re-distributed (Maxwell, 2013).

In conclusion, recent neuroscience advances are demonstrating first, that interconnected brains can perform multiple computational tasks, allowing for the appearance of a world wide brainet; and second, that such brainet could use blockchains to prevent attacks from non biological entities.

References
1 – Robotics: Ethics of artificial intelligence, Nature 2015, 28th May
2-Nicolelis 2011 Beyond boundaries
3-Siegelman 1995 Science
4-Pais-Vieira et al., 2013 BBI paper
5-Deadwiler et al., 2014?
6-Ifft et al 2014 sfn Abstract with monkey brainet (Arjuns paper?)
7-Rao et al., 2014
8-Mauro K, 2015 Grand Scholars Challenge
9-Kurzweill in Neuman 1958 Singularity
10-Nakamoto 2008

11-https://www.whitehouse.gov/blog/2014/10/09/brain-initiative-and-grand-challenge-
scholars

12-Maxwell, 2013 – https://bitcointalk.org/index.php?topic=279249
13-Future of life- : http://futureoflife.org/static/data/documents/research_priorities.pdf

Services in a Brain-to-Brain Internet

The below article is intended for a very futuristic use case, involving implants. You can study actual, ready BTB services at our github.

Future plans: Imagine a service called YouLive. Rather than sharing user-uploaded videos (as YouTube does), YouLive would share user-uploaded experiences. In such a system of shared experience, care must be taken to ensure that a user’s downloaded experience does not exceed the length of the recording, and that the downloaded experience is of the quality expected by the receiver. A user’s brain must be fundamentally protected from becoming ‘trapped’ in another (potentially ill-acting) user’s experience.

On a POC blockchain, safety is ensured by publicly-announcing connections between humans and services. The value or script of the transaction would be signify the length of time a connection was assumed to be valid. A human would confirm a connection time after receiving a service’s proposal. Once a connection was terminated, a human would then send a second transaction to signify they had been disconnected. The publicized initiation and termination of a connection is crucial to the safety of the network. If a connection was not terminated after the specified time, users on the network would notice on the blockchain, and could remove the ill-acting service from the network, or simply leave the network themselves. Once again, the decentralized nature of blockchains offer a safer means for these broadcasts than a centralized server, the latter of which can be hacked. Thus, a blockchain is absolutely necessary as a foundation for brain-to-brain security.

In the bitcoin protocol, a small number of ’emergency broadcast keys’ are given to core developers to alert users in the case of network failure. These keys could be used to automatically disconnect users in the event a number of downloaded thoughts or service connections did not receive their termination transactions.

In the case of human-service handshakes, a minuscule digital currency fee would be incurred to allot for the cost of blockchain data storage, as well as to help support the centralized structure hosting the experiences. If the service were decentralized (which it should be), a portion of the digital currency fee could be sent to the user who uploaded the experience, or whatever other feature the service offered. This fee would be necessary as an incurred cost for network spam, and relatedly, would ensure human nodes with a poor level of trust could still have their human-service transactions posted on the blockchain.

The protocol for a brain-to-brain internet outlined here is one that is discontinuous. Downloaded thoughts and service connections would only be utilized for an agreed-upon length of time. Code signing (Kiehtreiber and Brouwer [2006]) would ensure that downloaded thoughts were unaltered, similar to how file downloads on the internet are secured. Continuous, live stimulation of neurons will probably never be a safe mechanism for brain-to-brain communication, though refreshable service connections may allow for continued data sharing.

 

Reference: proof-of-cognition-implants , published May 2015. Disclaimer: Project Oblio’s mechanism does not rely on brain implants, but the mechanisms of action are the same. An early version of the paper provably exists in bitcoin address 13eeMVU5fXNfZdoBk5z4fEAbgSH9MawQ6H.

Transcranial Direct/Alternating Current Stimulation in Boosting Memory

Noninvasive brain stimulation techniques are gaining attention due to their safety in modulating brain dynamics. Promising applications include treatment of various central nervous system diseases and improvement in cognitive functions. Transcranial electrical stimulation provides noninvasive brain modulation using direct current, alternating current and random noise stimulation (Paulus, 2011).

A weak direct electrical current is applied through the scalp using two or more electrodes in the  transcranial direct electrical stimulation tDCS technique. This induces brain excitability through cathodal hyperpolarization and anodal depolarization (Paulus, 2011). The induced effects depend on polarity, duration and intensity of electrical stimulation.

Transcranial alternating current stimulation is same as direct current in case of low intensity and electrodes. But it uses sinusoidal current through the scalp for electrical stimulation (Woods 2011). Several studies have been carried out that showed that transcranial direct/alternating current has a positive impact on motor, memory, perception and cognitive functions.

Cognitive processes include perception, memory, learning, and long-term memory formation. Induction of transcranial electrical brain stimulation enhances the cognitive functions. Direct current stimulation occurs through spontaneous cortical activity and alternating current modulate cognition by interfering with the oscillations of cortical networks (Kuo, 2012)

Transcranial stimulation of the brain through weak direct current induction serves a non-invasive and painless technique (Nitsche, 2000). In their study, induction of direct current through the scalp for modulating motor cortex excitation, showed up to 40% of the excitation changes that last for several minutes after end of stimulation. Stimulation was achieved by membrane polarization inducing anodal stimulation and inhibiting cathodal stimulation.

Weak direct current induction leads to cerebral excitability. Fregni et al (2005) evaluated “the effect of anodal stimulation of dorsolateral prefrontal cortex (DLPFC) on working memory”. A letter-based working memory task was performed by fifteen individuals during anodal stimulation of DLPFC. Out of these seven performed the same task but with cathodal stimulation. Results showed the increased performance of individuals with anodal stimulation.

Another study was carried out to investigate the association of slow oscillations on memory during sleep. Induction of transcranial slow oscillations of 0.75 Hz in early sleep increased the retention of declarative memory in healthy subjects and also improved the slow wave sleep and slow spindle activity in the frontal cortex. Stimulation of the brain by 5Hz oscillations during rapid eye movement sleep had no effect on declarative memory (Marshall, 2006). Based on this, another study was carried out to rule out the effect of slow oscillations during waking on brain and memory encoding. It was concluded that the effect of oscillation and memory depend on brain state, as when the awake brain transmitted stimulation by responding to oscillations and facilitated encoding. Transcranial oscillations didn’t improve memory when applied after learning, while it showed enhanced encoding of hippocampus dependent memory when induced during the process of learning (Kirov, 2009)

Transcranial alternating current stimulations have an enhanced effect on human cognitive functions. Antonenko et al (2016) conducted research on young and older healthy individuals. Transcranial alternating current of 6Hz was applied to the brain for 20 minutes during a language learning process. The results were in support of the evidence that alternating current improves human cognition through direct stimulation of task-related brain oscillations.

 

References

Antonenko D, Miriam Faxel, Ulrike Grittner,Michal Lavidor and Agnes Flöel, 2016. “Effects of Transcranial Alternating Current Stimulation on Cognitive Functions in Healthy Young and Older Adults”. Neural Plast: 4274127. Doi: 10.1155/2016/4274127. PMCID: PMC4889859. PMID: 27298740

Fregni, F, Boggio, P.S, Nitsche, M. et al, 2005. “Anodal transcranial direct current stimulation of prefrontal cortex enhances working memory”. Exp Brain Res: 166: 23. https://doi.org/10.1007/s00221-005-2334-6

Kirov R, Carsten Weiss, Hartwig R. Siebner, Jan Born, and Lisa Marshall, 2009. “Slow oscillation electrical brain stimulation during waking promotes EEG theta activity and memory encoding”. PNAS; September 8, 2009. 106 (36) 15460-15465; https://doi.org/10.1073/pnas.0904438106

Kuo M F, Michael A. Nitsche, 2012. “Effects of Transcranial Electrical Stimulation on Cognition”. Clinical EEG and Neuroscience, Volume: 43, issue: 3, page(s): 192-199. https://doi.org/10.1177/1550059412444975

Marshall, L., Helgadóttir, H., Mölle, M., and Born, J. (2006). “Boosting slow oscillations during sleep potentiates memory”. Nature 444(7119):610-3. doi: 10.1038/nature05278

Nitsche MA, Paulus W, 2000. “Excitability changes induced in the human motor cortex by weak transcranial direct current stimulation”. J Physiol. 2000 Sep 15; 527 Pt 3:633-9. PMID: 10990547 PMCID: PMC2270099

Paulus W, “Transcranial electrical stimulation (tES – tDCS; tRNS, tACS) methods”.Neuropsychol Rehabil. 2011 Oct; 21(5):602-17. Doi: 10.1080/09602011.2011.557292. Epub 2011 Aug 5.

Transcranial Stimulation In the Treatment of Depression And Mood Improvement

The largest study conducted so far with respect to the application of transcranial direct and alternating current stimulation in the treatment of depression was published by Brunoni. The author asserted that he made a controlled trial with over 120 patients suffering from depression. This resulted in a factorial study in which the patients subjected randomly to receive active tDCS and serum sertraline/placebo exhibited significant symptoms as compared to the patients that were given active tDCS and also in combination with sertraline.

As a result, further randomized clinical trials that aim to evaluate the clinical efficacy of tDCS in depression are being performed worldwide.

Here are some of the benefits of using transcranial direct and alternating current stimulation in the treatment of depression:

Transcranial alternating and direct current stimulation improves memory

The use of theta waves on the left parietal of a depressed patient helps to increase the working memory as well as factual memory as shown in 12 ADHD children. It is interesting to note that this only works when the theta waves are in the synchronization phase. Similar research conducted on 12 healthy female children reveals that it increases the memory confidence.

Transcranial direct and alternating current stimulation changes the brain waves thereby fighting depression

Three different studies were conducted on humans for about 20 minutes and it was discovered that transcranial direct and alternating current significantly increased the brain power for about 30 minutes in the indicated wavelength range.
Conclusion

Transcranial direct and alternating current stimulation is an appealing treatment for depression as a result of its relative safety and efficacy profiles attached to the fact of its relative inexpensiveness. TDCS have also been found to have tangible anti-depressant effects. It is also considered a promising therapy because of its minimally invasive nature and its benign relative adverse effects.

However, further research is required to examine the utility of transcranial direct and alternating current stimulation as the first treatment to think about in more severe forms of depression. Presently, it seems crucial to consider transcranial direct and alternating current stimulation as a treatment for patients with a mild level of depression without resistance to treatment. It may be just as effective to make use of it in enhancing the first kind of response rates when combined with pharmacotherapy and psychotherapy.

 

References

1Antal, A., Boros, K., Poreisz, C., Chaieb, L., Terney, D., & Paulus, W. (2008). Comparatively weak after-effects of transcranial alternating current stimulation (tACS) on cortical excitability in humans. Brain stimulation, 1(2), 97-105.

http://www.psychiatrictimes.com/neuropsychiatry/current-status-transcranial-direct-current-stimulation-treatment-depression/page/0/1

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4369553/

https://www.sciencedirect.com/science/article/pii/S0014488609001290

https://www.cambridge.org/core/journals/psychological-medicine/article/transcranial-direct-current-stimulation-in-the-treatment-of-major-depression-a-metaanalysis/96254C1048E1706414248C27C4E9BCA5#

Brainwaves as a Future-Resistant Biometric: Human-Detection, Identification, and Authorization

Brainwaves are the only biometric still untapped by governenments and corporations that possess fluid properties; Human-Detection, Identification, and Authorization. They are also a source of wealth.

“Human detection” is the act of proving whether or not an internet user is a robot, or a human. Google’s ReCaptcha2 (“I’m not a Robot”) is very convenient; It only takes about 10 seconds to complete. Unfortunately, the tasks required by Google’s reCaptcha are problematic for the following reasons:

  • These tasks are not future-proof. Eventually, A.I. will be smart enough to pass any image recognition task.
  • As Google trains their algorithms, they become the only ones with the algorithms capable of bypassing their own human detection protocols. They can then lease these algorithms to generate fake accounts. These fake accounts, paid for by a wealthy person or persons, or just a hacker, can be used to create a false sense of consensus on a discussion board regarding a government, product, or cultural content.

Project Oblio possesses a network component that is decidedly human-only. Transactions on this network can be decidedly human-only, as the network revolves around the idea that the most resilient form of human detection derives from the power of the human brain. Signals emanating from the brain and muscle areas around them are a cheap and efficient form of human detection, one that is vastly more future-proof that that currently implemented by Google.

Each of us is born with a multi-billion dollar supercomputer, capable of generating outputs immeasurably more complex than that capable of being understood by an A.I. For example, predicting whether a human will find something funny is much harder for A.I. than it would originally seem, even with plenty of data. This humor response manifests itself in an EEG recording, a piece of data that can simultaneously be conveniently monetized to the user on a decentralized network.  Common recording parameters like the P300 (a measure of whether a human brain has detected something “surprising”) are easily elicited over an EEG and probably just as difficult for a computer to simulate. Any form of human detection becomes significantly stronger when you’re simultaneously recording outputs from billions of live, biological neurons.

When we combine this type of human detection with transcranial direct current stimulation (tDCS: an “input” method, as opposed to EEG, which primarily records a brain’s output), we may get even stronger, faster human detection. Considering that tDCS has also been shown to improve memory, concentration, and relieve depression, it would seem to be the perfect technique for improving both inward and outward human communication.

Although non-invasive brain-machine-interfaces require a lot of data for most tasks we’d like them to be useful for (such as control of virtual reality), they are tremendously accurate at identifying us (like a “fingerprint”, but more accurate) with comparatively minimal data. While the more desirable tasks often have only 60-80% accuracy, identification of human beings by brainwaves can typically achieve 99% accuracy with minimal data. This is evidence that each brain and its corresponding outputs vary tremendously from person to person, and that collecting a lot of data from a single person is just as valuable, if not more valuable, than comparing data across persons. If you need more evidence for the uniqueness of human brains, check out the Human Connectome Project.

So. At this point we have two new terms: Human detection, and identification.It’s important now to realize that not only do human brains have tremendous power to human detect us, but they can simultaneously perform these two steps as well. No other biometric available can do this simultaneously, and it’s a crucial enhancement in a decentralized network’s stability. Fingerprints and DNA are both static – once you have the image or the code, it can be copied and used to impersonate you. Voice data is as fluid as brainwave data and could theoretically be used to human detect, but it is not as good at identifying you and is also (nowadays) easily forgable (see Alexa, Siri, etc.).  If data were easily foregable, a person could use the same data to receive multiple rewards, spamming the system and making themselves very rich (that is, until the markets crash due to the flaw they exposed in the network).

The last step in the three-pronged approach is authorization. While human detection is like reCaptcha, and identification is like a username, authorization is like a password. Authorization is as simple as you thinking a “password thought” recognized by a machine learning algorithm. There are tons of papers already out there about this, so I’ll point you to Google Scholar for this topic.

Combining these three methods: identification, authentication, and human detection, into a single protocol creates a triad of network security not found in other biometrics or in purely digital security. This is  interchangeably called “proof-of-person”, “proof-of-cognition”, “proof-of-humanity”, and “proof-of-individuality”. Its consequences are described further here.