The Digital Dopamine

 2019-07-19 By: Elroy Yao

The InnoVEX 2019 forum had many speakers from companies around the world discussing 3 major topics: AI, blockchain, and IoT. While the three technologies have been buzzing around since several years ago, too many companies seem to focus too much on their individual utilization rather than their combined effect. Any technology when combined can effectively function as force multipliers of each other.


In relation to AI, blockchain, and IoT; what they can achieve together might possibly be the biggest world changing technology in the modern world. As Mr. Limas Lin stated, IoT devices or sensor modules can function as replacement for senses and AI systems as the brains; creating the most human digital analog, but where does blockchain play into the equation? Put simply, they might be the one thing that this digital human lacks: positive feedback.


Humans and the Ingrained Behavior


The human mind is considered the final frontier and the irony should be noted. How can something so near, dear, and ubiquitous be considered the final frontier? One theory that might be offered is that because humans are individuals and they give different responses to the same stimulus. However, despite all these different responses; humans in general are looking for positive feelings. This might be seen in how humans gravitate to the things they like consciously or subconsciously and steer away from things they do not. Humans also tend to repeat or do things that give them positive feelings or remove/ reduce negative feelings.


So far so psychological; so what does this have to do with the triumvirate of AI, Blockchain, and IoT? Well let’s look at the general consensus among AI researchers, developers, tech enthusiasts, and futurists: General AI is the holy grail of AI; the South Pole to their Amundsen; the Golden Fleece to their Jason. This general AI actually aims to mimic humans; the ultimate general intelligence. To be able to create something that mimics humans; it will be helpful to understand what makes humans tick: positive feelings or happiness.


While AI can be programmed and trained to include a certain form of reward for a job well done and aim to gather more of the reward; there have been some issues of cheating.So the reward system must also be reconfigured to make sure that the AI can only get it when they did a task properly and well instead of finding a work around that fulfills the criteria. With the characteristics of immutable, decentralized, secure, irreversible, and consensus based; blockchain might be the technology needed to secure the system and the reward itself. In other words, blockchain can be the reward (dopamine) and the security measure to ensure compliance.


Positive and Negative Reinforcement


Basing the idea from Mr. Dominik Schiener's speech in InnoVEX 2019; a form of cryptocurrency can be used to facilitate the exchange machine to machine communication. This means the application can be extended all the way from the proposed data/ information purchase to rewarding the machine when certain tasks or milestones are met. The system actually follows a carrot and stick training; if the tasks done were done correctly, they will be given the reward. On the other hand, the rewards will be “taken” if they failed to do the tasks correctly.


The training does not need to be limited to supervised learning; but developers can consider using a series of parameters as confirmation methods if the tasks are indeed done the way it should. One example from a recent case is when a quadrupedal robot needs to balance a ball on its back and walk without losing the ball. What happened in that case was the robot clung to the ball instead and then shambled forward. Perhaps a series of strict parameters can prevent the “cheating” when combined with positive and negative reinforcement; the idea of a carrot and stick.


Conclusion


The best practices for AI are still being compiled by everyone in the ecosystem; with numerous suggestions and counter suggestions proposed at any given time. While general AI remains the goal; a multi-use AI will be a more realistic goal that can be used as a milestone for developers and companies alike and at the same time delivers real value to the users in a nearer future.