Nick Bostrom's newest book Superintelligence illustrates a not-so-far future where machine intelligence surpasses human intelligence. Nick defines superintelligence as any intellect that greatly exceeds the cognitive performance of humans in virtually all domain of interest.
Pathways that superintelligence can occur
There are 5 pathways to superintelligence: Artificial intelligence, Whole brain emulation, Brain-computer interface, Biological cognition and Networks and organisation.
1. Artificial intelligence
The current Artificial Intelligence focuses on the idea of learning as a way of bootstrapping a simple system to human intelligence. This can be traced back to the idea of Alan Turing's child machine. 'A child machine' simulates the child's intelligence. When subjected to appropriate training, it will grow into adult intelligence.
A second possible way of artificial intelligence could be based on running genetic algorithms on sufficiently fast enough computers, thus achieve the results comparable to those of biological evolution. This process depends on both how much computing technology will advance over the next few decades and how much computing power is needed to run the algorithms. The computing power that is powerful enough to run such algorithms is severely out of reach on Earth.
The third possible means to achieve such artificial intelligence is using the human brain as a template for a machine intelligence. The different versions of this method is based on how closely they propose to imitate biological brain functions. However, to know when it will be achieved will be hard since it is difficult to predict the future discoveries of brain science.
An artificial intelligence must not need to resemble a human mind. It could be foreign indeed.
2. Whole brain emulation
Intelligent software would be produced by scanning and closely modelling the computational structure of a biological brain in whole brain emulation. In order to achieve whole brain emulation, it requires the accomplishment of the following steps:
- Firstly, a particular human brain's sufficiently detailed scan is created. Many scanning machines could work in parallel to process multiple brain slices simultaneously.
- Secondly, to reconstruct the three-dimensional neuronal network that implemented cognition in the original brain, a computer for automated image processing feeds the raw data from the scanners.
- Thirdly, the neurocomputational structure resulting from the previous step is implemented on sufficiently powerful computer. This would result in a digital reproduction of the original intellect. The emulated human mind now takes form as software on a computer.
However, whole brain emulation requires some advanced enabling technologies. There are three key prerequisites: (1) scanning: high-throughput microscopy with sufficient resolution and detection of relevant properties; (2) translation: automated image analysis to turn raw scanning data into an interpreted three-dimensional model of relevant neurocomputational elements; (3) simulation: hardware powerful enough to implement the resultant computational structure. There is a good reason to think that the requisite enabling technologies are attainable, though not in the near future.
Generally, whole brain emulation relies less on theoretical insight and more on technological capability than artificial intelligence. Compared to AI path to artificial intelligence, whole brain emulation is more likely to be preceded by clear omens since it relies more on concrete observable technologies and is not wholly based on theoretical insight. With greater confidence than for the AI path, the emulation path will not succeed in the near future ( approximately within the next fifteen years) because of several challenging preceding technologies which have not yet been developed.
To assess the feasibility of whole brain emulation, one must understand the criteria for success. There are 3 different levels of emulation success:
- A high fidelity emulation: having the full set of knowledge, skills, capacities and values of the emulated brain.
- A distorted emulation: whose makeups are significantly non-human in some way but which is mostly able to do the same intellectual labor as the emulated brain.
- A generic emulation: lacking the skills or memories that had been acquired by the emulated adult brain.
Compared with AI paths, the whole brain emulation path can be clearly indicated since it heavily relies on observable technologies.
3. Biological cognition
A third part to greater-than-current-human intelligence is to enhance the functioning of biological brains.
Though traditional methods of education and training can strengthen our individual cognitive capacities in various ways, biomedical enhancements could give bigger boosts. A person could improve dramatically upon cognitive functions by introducing some chemical into the brain of a healthy person.
The manipulations of our genetics will provide a wider and more powerful set of tools than biomedical enhancements. This means using selections at the level of embryos or gametes.
The derivation of viable sperm and eggs from embryonic stem cells would greatly enable the enhancement power of pre-implantation genetic screening. Fertile offsprings for mice or gamete-like cells in humans have been produced by the techniques for this. In this manner, it would be possible to accomplish 10 or more generations of selection in just a few years. There are some reservations around this technology. There is the unavoidable maturational lag while the selected embryos grow into adult human beings: at least 20 years before an enhanced child reaches full productivity, longer still before such children come to constitute a substantial segment of labour force. Even after the technology has been perfected, the adoption rate will start out slow. Some countries might ban its use altogether on moral or religious grounds.
Other potential biotechnological techniques might also be appropriate. Human reproductive cloning could be used to replicate the genome of the most exceptionally talented individuals. Uptake will be limited because of the preference of most prospective parents to be biologically related to the children; however, practice could have non-negligible impact because (1) even a relatively small increase in the number of exceptionally talented people can have a significant effect; (2) there is a possibility that some state would start on a larger scale eugenics program.
In summary, firstly, at least weak forms of superintelligence are achievable by means of biotechnological enhancements. Secondly, the feasibility of cognitively enhanced humans adds to the possibility that advanced forms of machine intelligence are feasible, because if we were unable to produce machine intelligence, cognitively enhanced humans might still be able to create it. Thirdly, we have to take into account the probable emergence of genetically enhanced populations with the magnitude of enhancement escalating quickly over subsequent decades when we consider scenarios throughout the second half of this century and beyond.
4. Brain-computer interfaces
Direct brain-computer interfaces could enable humans to exploit the fortes of digital computing and the resulting hybrid system can radically outperform the unaugmented brains. However, it is unlikely that such interfaces will be widely used as enhancements any time soon. Firstly, there are significant risks of medical complications when implanting electrodes in the brain. The second reason is the enhancement is likely to be far more difficult than therapy. For example, patients suffering from paralysis might benefit from an implant that replaces their severed nerves or activates spinal motion pattern generators, patients who are deaf or blind might benefit from artificial cochlear and retinas. These can be achieved by using a regular motor and sensory organs to interact with computers located outside our bodies.
Despite these reservations, the cyborg route toward cognitive enhancement is not without promise. The feasibility of a neural prosthesis can enhance performance in a simple working-memory task. There is hope that the cyborg route to the brain would over time learn an effective mapping between its own internal cognitive states and the inputs it receives from, or the outputs accepted by, the device. Then the implant itself will not need to be intelligent but the brain will intelligently adapt to the interface.
5. Networks and organisations
Another possible path to superintelligence is by gradually enhancing networks and organisations which link individual human minds with one another and with various artifacts and bots. The idea is that some system composed of individuals networked and organised might attain a form of superintelligence.
The technological and institutional innovations could contribute to the growth of our collective intelligence are abundant. Subsidised prediction markets might foster truth-seeking norms and improve forecasting on contentious scientific and social issues. Lie detectors prove feasible could reduce the magnitude of human deception in human affairs. Self-deception detectors will help reduce the forms of deceptions.
Growth in collective intelligence may also come from more general organisational and economic improvements, and from enlarging the fraction of the world's population that is educated, digitally connected, and integrated into global intellectual culture.
Summary
In conclusion, there are many paths leading to superintelligence, which should increase our confidence that we will eventually get there. If one path is blocked, we can still progress.
Enhanced biological or organisational intelligence would accelerate scientific and technological improvements, potentially hastening the emergence of more radical forms of intelligence enhancement such as whole brain emulation and AI.
Multiple paths do not mean multiple destinations. Cognitively enhanced individuals will further accelerate the machine intelligence's growth. However, the path taken to get there can make a difference to the eventual outcomes. Cognitively enhanced & organised intelligence can make machine intelligence less risky.
True form of intelligence will start with AI path first. The whole brain emulation is somewhat likely to be the quickest way to get to superintelligence.
Biological cognitive enhancements are obviously feasible, especially when it is based on genetic selection. This will further our confidence in the conceivable idea of machine intelligence since cognitively enhanced engineers and scientists can create machine intelligence much more likely than the normal human brains. Brain-computer interface is unlikely as a means of superintelligence. Improvements in networks and organisations might be weakly superintelligent forms, however, they will play an important role as the cognitive enhancements to achieve machine intelligence to a lesser extent.