Among the multiple viewpoints made with regard to AI, it looks like that the progress of AI is being stuck between fear and threats assumed by a particular set and doubt among some thinking this is just a hype. However, the fact would remain that the AI machine would be harmless like any other machine around us and will certainly be more intelligent than humans expect it to be.
This post attempts to put together simple points to logically explain why an unsupervised machines will prove to be harmless and deliver towards the benefit of the human race.
First, Lets rationalize why machines will not be a threat to humans
#1 Machine Processing is different from Biological Processing for the fact that machine need not undergo the process of converting responses to electro-chemical energy. This process is evident in biological systems that they trigger hormonal discharges in order to transfer energy packets to the target location to execute the decision, experiencing an emotional state. As machines can execute decisions using electrical energy, they can skip emotional states and avoid all associated errors.
Unlike biological forms which add up varying emotional weights creates an additional bias, which could lead to errors. The machine uses uniform allocation of weights for every machine-driven stimulus-response interaction, avoiding pattern irregularities, that is collected among biological forms due to ligand interactions and interaction potential. Hence, AI machines have no scope to demonstrate fear, anxiety, hatred, compassion or love.
#2 Today, The fear of AI is driven by assumptions. Due to limited understanding or due to the influences of movies or books, most people believe that AI can be so powerful that they might take control of humans and cause insecurities and threats to human existence
In order to understand this better, let’s decode fear because it appears that fear and intelligence are inversely proportional to each other. The assumption that machines are a threat arises from our primal response of fear. Fear can be explained as a repulsive behavior because of minimum energy reaction paths, which is considered as the most basic default response found in all beings.
Due to lack of or fewer response patterns to handle the threat, we would switch to the pattern of distancing from the object)in order to regain our composure, which can be explained as regaining our normalcy states(equilibrium) . When patterns are repetitively collected from exposure/experience, newer response patterns are developed which would present a more optimal response to maintain equilibrium. This would allow the being to move away from the state of fear, giving them more control to overcome the state.
A simple example can be witnessed between two bike riders. A novice rider who’s riding patterns are fewer in number due to lesser exposure is more fearful than the other expert rider who’s riding patterns are vast and diversified.
You can see that the patterns(data) play a crucial role in managing threats. The more we expose ourselves to the threat, we are learning a new pattern that helps us manage these threats.
In the context of information flow, threats can be described as an erroneous decision backed by inaccurate intelligence derived from incomplete knowledge, formulated from heuristic relationships(short term goals) and restricted to collecting limited data parameters.
We can also witness that in the evolution timeline, we have seen fear (repulsive behavior) as a common basic trait among all animal species. Intelligence, on the other hand, has grown throughout the timeline. The exponential growth in intelligence can be witnessed by broader thinking (incorporating every influencing data parameter), which can result in answers to challenges and hence, overcome fear.
So you could also say that more the intelligence about a certain aspect, the lesser the chances of experiencing fear around it. I feel that the day we realize that AI machines are nothing but hardware which by way of systematic data processes can mimic human like intelligent behavior, we will be in a better position to ward off the fear.
#3 Even an unsupervised completely autonomous machine is a piece of machinery code
The other basic aspect that even the super intelligent unsupervised machine, which can learn and perform autonomously, works on a basic set of code implemented by humans, which would mean that it gives us great control to manage super-intelligent machines.
Technically, it would mean that through the interface, users can query for objects and understand the various aspects, relationships, and their weights to predict what might be the possible next actions of the machine. Users will be able to manage weight threshold so we know that the super intelligent machine can be completely supervised, in the case of an unwarranted scenario.
The worry that there might be certain people who might put an evil program into the machine, should consider the fact that machines are not emotional. It has to be a command that it needs to executed in order to kill, in which case it would seem that some of the patterns have been configured with decision presets, which makes it supervised.
Supervised machines will always be way less intelligent when compared to unsupervised machines. The very thought that it has to classify on its own when it experiences the external environment, cannot be handled by supervised machines due to boundaries set by the program owner.
# Machines do not need our money
The machine has no other intent apart from amassing knowledge and intelligence so it can perform accurately. It is not looking for the same resources as us like food, money, land to become a competitive threat. It would be nothing but an intelligent machine where humans can dip into it to make logical decisions and progress faster .
As far as the job threat goes, the Universal Basic Income is a good strategy. Due to automation, businesses might reap more profits which can be collected back as taxes to pump it back into the UBI scheme, creating a balance in the income groups. Due to automation, prices are likely to fall and make way for a more affordable future across the world.
Humans, with time in hand, can pursue what they really like to do and live a more balanced life to achieve longevity. Could we see a serious uprising in the quality of art or adventure that the human discovers new possibilities in an environment where intelligence is so accessible?
With that rationale, I would put aside the fear of AI and look forward to these intelligent forms.
The second part of our deliberation would be to substantiate how machines can be extremely intelligent beyond human expectations
#1. A Human is also a machine. With the help of scientific evidence, we are now aware that cellular forms are nothing but a optimized organization of molecular machinery, which logically follows a specific pathway to conduct a response to the triggered stimulus. Unlike the solid machine, these cellular machines resorts to electro-chemical pathways to conduct itself.
The stimuli-response pattern is common in all processing forms and the data processing pathway in the human brain is no different. We have also learnt that the evolved mammalian brain (humans) exhibits three tier response system based on available time. Reptilian response (coming from repetitive patterns with high probability weights), Limbic responses (coming from macro weights and primary emotional bias) and responses from Neocortex which involves higher analysis and over larger distribution of parameters.
Between these response types, lies the experience of consciousness or awareness. It is simply the speed at which responses are delivered that we classify the states as conscious or subconscious. When the body renders a reptilian response without our voluntary attempt of computation and selection, we consider it as subconscious decisions. In case of a voluntary response(conscious), we experience the response paths all the way to the neocortex and during this time, which can be termed as the state of being aware.
Unlike the human counterparts, the machines have a single response system mimicking the neocortex responses. We can say that the machine are always conscious (aware) and never experiences the state of subconscious. There might be scenarios where the machine might be able to render highly analyzed responses at the speed of reptilian responses as it can manage its time better by way of predicting the scenario.
#2. Machines can demonstrate reasoning and abstraction just like humans.
From the available logical theories on how human brain process, we see that the brain uses a simple methodology of detecting similarities and differences in a linear pattern and selects a high energy path to deliver a response. Information is processed at certain areas of the human brain to create a network of long linear patterns based on their firing in unison (Hebbian Theory).
These pattern is good enough to detect the differences and learn to make sense of the outside world. Behavior like perception, learning, reasoning, abstraction are drawn out from these patterns. Documentation on how different aspects of pattern recognition are directed to exhibit these behaviors explains how a machine can easily demonstrate complex human-like intelligent behavior.
#3 Machines can make far accurate decisions than humans if the confirmation threshold is set high.
Machines can make erroneous decisions just like humans if the machine has confirmed to the fact with little data and has not reasoned enough. Most humans confirm patterns as facts either by experiencing it couple of times or agreeing to information gathered from third party sources. Most decisions coming from these confirmed facts lacks deeper analysis (reasoning) which can lead to inaccurate expectations and and erroneous results. You would see such instances in software programs where it is purely an instance of heuristic data computation over macro inputs leading to wrong results.
Such confirmation logic adds up to the scare that the machine can learn wrong patterns and execute them causing risks to humans. It is important to raise the confirmation threshold to help the machines make more accurate decisions. Machines can be set with a higher confirmation thresholds, which means that the machine will act up on a pattern only when it records 50 repetitive exact patterns (all along its complete network). This would allow the machine to reason enough before making any intelligent decision.
In my view, the machine would do better in terms of being a greater judge or in terms of reaction time or even risking itself to place a human in a more predictable and secure environment
#4. Why we are not there yet?
The current AI experiments takes a more top’s down approach and tries it solve heuristically. The approach to deliver a safe dependable AI requires us to move away from silo solutions and work on an integrated solution (more like the brain). At the end of the day, the job of AI is to mimic the information processing pathway exhibited by the human brain.
The current approach works in silos such that the language works independently of the visual database, while in the brain, both data patterns are integrated. Most NLP applications are restricted to the keywords and encounter restrictions without any visual cues, making it hard for machines to explain an abstract of a concept or explain a story or raise a question based on what it sees.
Likewise, we also see that there seems to be a gap in understanding the optimal data structure required to make the learning models work seamlessly. There is more stress to understand which model works best without optimizing the semantic relationships of the data itself.
Tackling these integration challenge holds way to the more human like autonomous machine demonstrating Strong AI.
With the speed of which AI is progressing, we are couple of years away to see the first autonomous machines among us and I would bet that by this time we would be more encouraging than fearful, than we today are.