Deep learning. Risks of fraud with AI.
Up until recently, traditional machine learning (if it is possible to call traditional an industry that did not exist before 1959) had a benefit of full and complete introspection. This means that machines were making decisions and scientists then could track the entire path of how a machine would arrive at a decision. Sometimes, this can be very useful. For example, in law, it is important to explain how a new decision links to old decisions and precedents. In making a decision about approving or denying a loan, a financial institution needs by law to explain why it has made the decision it has made.
However, this very advantage of being able to track the full path was exactly what was preventing the machines from working with unclassified data.
Introduction to deep learning
With the subset of machine learning called deep leaning, it became possible to work with large amounts of unclassified data. With deep learning, machines can not only establish correlations between, say, images and names of people as a part of a facial recognition process but also establish correlations between past events and future events.
The benefit that a machine has over a human in this regard is that the machine doesn’t care whether something has happened in the past, is happening in the present, or will happen in the future. For a machine, a prediction is just another label or outcome that is no different from all the other labels or outcomes. Machine learning does not care about the times of events. Given a set of data, a machine can analyze the data and make decisions based on the data.
Some of the obvious uses of such predictions would be device and hardware breakdowns in data storage, manufacturing and cargo industries, health breakdowns of people, customer churn, meaning predictions about when and why customers would leave, and employee turnover.
Many of these predictions can have very important consequences. For example, if a business knows when and why customers are likely to leave, it can take action to try and change that, such as sending customers a gift, giving them a discount, or try to alter the flow of events in a different way.
The better people and organizations can predict what happens, the better they can prevent the negative critical events that are likely to occur.
The potential for fraud and risks associated with machine learning
Obviously, having information about future events can be highly lucrative and this means a lot of risks.
The first risk is the security of information and protecting the information from attackers. As the Equifax hack of 2017, eBay hack of 2014, Yahoo hack of 2013 and others show, no company today is immune to cyber attacks. One of the reasons for this is the centralized nature of storage and operations of modern technology companies. Typically, there would be a server or a group of servers that hackers can try and penetrate. Compared to centralized storage of information, blockchain networks are highly secure. This means that blockchain technology is perfect for storing highly sensitive information that needs protection.
For example, to access a wallet on the Bitcoin network, you would need a pair of keys, a private key, and a personal key. Without the private key, you will not be able to access the wallet or send funds. On the Bitcoin network, the information is about financial transactions, but a different blockchain network could store different information, such as information about medical records or location of cargo containers. The security and principles behind the technology, however, would remain the same in that without a pair of keys a user will not be able to access information or participate in any kind of information exchange, which is what Bitcoin transactions are really about.
The second risk of dealing with highly sensitive and valuable information created by artificial intelligence is the validity of information and the validity of claims about such information.
Obviously, as artificial intelligence becomes more and more widespread, there will be companies that would claim to have the “secret sauce” and to be making the best predictions about markets, weather conditions, health conditions, or something else. No matter how good of a job the government does to protect the citizens, there will be players in the market trying to defraud the public. With predictions, an obvious choice of a malicious company would be to try and alter past predictions. With blockchain, it is impossible to remove or alter information that is already on the blockchain. This means that if a company were to store certain data on a blockchain, users would know that the data has not been altered. In addition to this, blockchain networks are decentralized, meaning that multiple computers can have copies of the full blockchain and even if an attacker were to alter multiple copies, the network would be able to restore itself to the original condition even if there’s just one copy of the original blockchain remaining.