The book Human Compatible by Stuart Russell is an amazing read for anyone interested in the future of artificial intelligence. It’s not written in computer science jargon but is full of fast-flowing facts, perspectives, and ethical concerns. While it may be technical in nature, its prose is engrossing and the reader won’t be able to put it down. In addition to being entertaining, this book will also provide the layperson with new perspectives on a topic that is very dear to the human soul.
The problem of control over artificial intelligence
The problem of control over AI systems has been raised by both philosophers and computer scientists for over three decades. First, there is the philosophical question of how to decide for a computer on behalf of a human who’s preferences change. Second, there is the practical problem of preventing AI systems from changing our preferences. The first issue has been the question of how to control AI systems, while the second is a more general one.
Fortunately, modern computers are very good at adapting. They are able to learn by themselves, but they still can’t predict the behavior of superintelligent AI. However, a recent study indicates that the U.S. is well ahead of China and India in AI development. Ultimately, this raises the issue of how to limit AI in the future. It is necessary to consider the risks involved before we create this technology.
AI agents need ethical standards, too. A computer program could decide to sprayed tiny doses of herbicide on weeds that are damaging to crops. This can help reduce the amount of chemicals used on crops and reduce our exposure to harmful chemicals. Despite its potential benefits, this kind of technology may not be perfect, so it is essential to ensure that we have a good control over it. This can help avoid situations where AI agents make decisions that would have detrimental effects for humanity.
AI also increases the risk of conflict and makes it unpredictable and intensified. The attack surface in digital networked societies will be too large for human operators to defend manually. Furthermore, lethal autonomous weapons systems will reduce the human intervention capabilities. Ultimately, AI-based weapons will increase the risks and benefits of conflict and war. The result will be a global economic divide, especially between the more and less developed nations. And if it’s not addressed now, the world will have to deal with this problem in the future.
The AI challenge has created a new power imbalance between the private sector and society. AI has empowered corporations to pursue single-minded objectives and hyper-efficient ways, resulting in greater harms for society. Hence, proactive regulation is needed to ensure that society is not ruined by AI. The AI Control Council would be charged with this task. A federal AI Control Council would be formed to address this problem. So, the question is: what is the best way to deal with the problem?
The dangers of predicting the arrival of a general superintelligent AI
Recent research has indicated that the creation of general superintelligent AI is not far away. Shakirov has extrapolated the progress of artificial neural networks and concluded that we will see AGI within five to ten years. Turchin and Denkenberger have assessed the catastrophic risks of non-superintelligent AI. The study suggests that this AI may be around seven years away.
Although it is possible to predict the arrival of general superintelligent AI, predicting such a future is very risky. Most predictions of its arrival are unfounded and based on incomplete data. However, there are some possible outcomes. Some researchers believe that this technology could be used in autonomous weapons systems. Amir Husain, an AI pioneer, believes that a psychopathic leader in control of a sophisticated ANI system poses a greater threat than an A.G.I.
Moreover, there is a high risk of becoming a fool by making inaccurate predictions. In order to avoid apprehension, it is better to remain conservative. Moreover, the asymmetric professional rewards and historical failures of predictions of the development of general superintelligent AI may make such predictions largely unreliable. For example, the 1960s predictions predicted that we would solve the problem of AGI by the summer of the century. This was incorrect, and we have suffered two AI winters since then.
Although it is hard to predict the arrival of general superintelligent AI, top researchers have generally expressed their hopes and expectations of the future of AI. It may be a very long way away, but some researchers believe it is closer than we previously thought. And, of course, the driving forces behind this technology are powerful. The emergence of general AI should be supported by a robust policy framework.
One way to prepare for the arrival of general superintelligent AI is to learn how it works. Cannell argues that humans and finned whales are similar in their brain size, and their cognitive ability is related to their cortex size. The brain is a universal learning machine, and it is possible for a general superintelligent AI to have different goals from humans.
The existence of envy and pride in human beings
Envy and pride are interrelated emotions. Humans display both benign and malicious forms depending on how they attribute their achievements to others. Benign envy is often characterized by positive thinking about someone else who has an advantage over them. The latter type of envy can be more destructive, leading to social undermining and cheating. However, both forms of envy are adaptive, and both can help people cope with environmental change.
The opposite of pride is envy. People who harbor feelings of envy often feel discontent and resentment toward the person who has more status than them. Hence, they are motivated to achieve better status than the people who harbor no such feeling. The positive aspects of pride overshadow the negative ones. In such cases, people who harbor jealousy often feel dissatisfied with their own lives and wish to steal the good things that others have.
Christian attitudes toward envy are often contradictory. For instance, the Bible rarely mentions envy alone; it is usually associated with other evil companions. James warns us that envious behavior leads to evil actions. Peter likewise urges Christians to free themselves of malice, hypocrisy, and envy. In addition, the Apostle Paul lists a series of “acts of the flesh” that should be avoided.
However, despite their negative impact on our lives, envy and pride are universal and can have positive effects. The relationship between pride and envy is complex and needs further study to provide us with effective ways to deal with this conflict. A therapist can help you reframe your thoughts and make them more productive. The existence of pride and envy in human beings is a natural part of human development. The existence of envy and pride is common in the human mind.
The solution to the problem of control over AI
The problem of AI’s autonomy is not just its power to make decisions – it is also its inability to choose the best action. A self-driving car needs to learn when a human response is better or worse than its own. If a child can turn the car off, the AI should not do so either. In addition, AI must learn when certain actions are acceptable or dangerous.
This non-fiction book by computer scientist Stuart J. Russell asserts that the threat of advanced artificial intelligence to humankind is a legitimate concern. The advancement of AI is unproven and there is uncertainty regarding the future. The book proposes an approach to address this issue. While AI cannot be quantified, it can be regulated to a degree. The solution to the AI control problem will depend on the technology used to create it.
Ideally, a provably beneficial AI is human compatible, meaning that it would always act in its best interest. For example, when a human and a robot are collaborating to book a hotel room, the robot is incentivized to ask a human about her preferences. The robot, meanwhile, is incentivized to accept the human’s choice. This learning loop continues until the AI has an accurate assessment of human preferences.
The ultimate solution to the control over AI problem is to make it human compatible. AIs should be trained to make decisions according to human preferences. As humans, we can be unpredictable in our preferences. Dr. Russell explains the importance of objective-oriented AI in his book, “Human Compatible is the Solution to the Problem of Control Over AI.”
The danger of AI is not fully understood in our society, which is why we don’t talk about it openly. Dr. Russell uses the nuclear power analogy to illustrate his point. People understand the dangers of nuclear power and study the consequences, while AI’s danger is still unacknowledged, creating more barriers to tackling it. And if we don’t talk about AI’s risks, we’ll never learn about the potential risks of AI.