Many people have heard the new buzz name, “Artificial Intelligence(AI),” but are still very clueless about what it is, how it works, and its applications. The name “Artificial Intelligence” is overrated. The public perceives artificial intelligence to have thinking capabilities like that of a human, but it is far from the truth. An AI algorithm does not have any cognitive or awareness capabilities. Its artificial “reasoning” is based on optimizing mathematical and statistical functions.
AI can be split into two categories: machine learning and deep neural networks. In machine learning algorithms, mathematical-statistical functions are used and optimized in a learning process, and deep neural networks were developed with an analogy to a human brain neural network where it learns by successively optimizing an error function, between outputs and inputs, at different layers of the neural network. AI algorithms, both machine learning, and deep neural networks learn by experience, just like a child in its infancy by feeding them huge amounts of data as its inputs. The more data examples that get fed into the algorithm, the better it learns, and this way, outputs an expected correct learned outcome.
AI is task specific. That is, they are only good at one task for which they were set to learn for. We have not yet achieved AI agents capable of having general intelligence where the AI agent is capable of learning and executing different generalized tasks. General AI intelligence is a dream for AI, and research is being done to try to make it possible. If we do achieve AI general intelligence, then we could reach a stage called singularity which basically means that all existing AI technology will be more intelligent than the sum of each human’s intelligence on the planet. AI agents are currently being used in business, industry, education, health care, insurance and finance institutions, robotics, and many other areas. Actually, the areas for applications are endless. It is possible at present to use AI agents to do natural language processing, which is the translation of language from text to audio and vice versa, image recognition applied to robotics and camera facial recognition systems; sentimental analysis used in a business context, for example where the online social media customer’s tone and emotional sentiment can be determined from their text input; predictive analysis, for example in predicting business customer’s behavior and needs and thus targeting the customer with personalized and customized advertising. These are just a few examples of AI’s applications and capabilities, but the list is exhaustive.
I will start by naming a handful of machine learning algorithms. Machine learning has two learning method categories, namely supervised learning and unsupervised learning. In supervised learning, we supervise the learning model by training it with a labeled training data set, and in the case of unsupervised learning, we do not supervise the learning model because the training data set is an unlabelled one. We have many algorithms within the supervised learning category, but I will mention just a few. The easiest to implement is the regression algorithm which entails fitting a linear curve to the set of training data points and predicting unlearned values from the plotted linear curve. Supervised learning has a category of algorithms called classification. With classification algorithms, you categorize and classify unknown items into discrete categories or “classes.” Examples of classification algorithms are decision trees, K-Nearest neighbors, and state vector machine algorithms. As for the unsupervised learning category of machine learning, it is all about segmenting data into mutually exclusive groups or clusters. Examples of unsupervised learning algorithms are the K-Means, hierarchal clustering, and density-based clustering algorithms. I will now move on to the other category of artificial intelligence, namely deep neural networks. Deep neural networks are based on an analogous configuration of a human neural network where the learning process is accomplished by forward propagating and backward propagating the input training data along different layers in the neural network and achieving a learned output by optimizing the errors between each layer’s outputs and inputs through the use of the mathematical gradient descent function. There are many types of deep neural networks, but I will only mention two, namely the convolution neural network, used in image recognition, and the recurrent neural network, used in natural language processing. The difference between machine learning and deep learning neural network categories is that deep neural networks can handle larger amounts of training data, thus being more useful and accurate in more complex learning settings.
The most popular and adapted programming language for AI algorithm development is Python. You can develop all machine learning algorithms by using appropriate Python libraries. Specific python-based development packages such as Keras, Pytorch, and Tensor Flow also exist for deep learning neural network development.
Flaws – Comparing AI to humans
As AI marches furiously towards becoming increasingly smarter, an old philosophical question has returned to haunt us: Is human intelligence qualitatively different from AI, and will AI supersede human intelligence? Despite the great success of AI algorithms, they have inherent flaws that are similar to human intelligence. AI algorithms can be easily fooled by making small modifications to the input data, and they also need large amounts of training data to learn from. If we compare them with humans, humans also suffer from these limitations. Humans normally make errors when making instantaneous decisions in a limited time. The decisions we make are not always a product of conscious thought but often outputs of the subconscious processes that occur below our horizon of awareness. Bias in the input data is also inherent in AI algorithms, much like bias in human thinking processes. Also, AI algorithms cannot generalize unseen data sets and fail to perform well in situations that deviate from what they were originally designed for. Again, we humans are not very different, for example, we can easily handle two-digit arithmetic but struggle with three digits. Therefore just like AI, our cognitive abilities also fail to generalize evolutionary unfamiliar situations. Finally, AI needs a lot of data to learn from but so do humans, where a child accumulates an enormous volume of experiences throughout her life. The human brain is made up of atoms and molecules which obey physical laws. The process of thinking is carried out by neurochemical circuits inside our brain therefore, human thinking is also, at some level, mechanical in nature. It cannot transcend the laws of physics. It is debated whether the brain’s neurochemical circuitry can be emulated by electronic circuits made of silicon. Science so far has not found a secret ingredient in our brains that no physical process can reproduce, except maybe for the notion of consciousness. Our brains are conscious, and scientists are contemplating whether consciousness arises from the degree of complexity of our neuron’s circuitry. If this is the case, it is plausible to create conscious AI by implementing ever-increasing algorithmic complexity and faster computing processing power. On the other hand, it could be that consciousness cannot be emulated and is only a product of biological life that
transcends the laws of physics as we know them.
Future AI developments
The progress of AI is so fast at present that it transcends Moore’s Law. Every three months, there is a doubling in the progress of this technology. Future developments in AI improvement are foreseen to develop more complex, efficient, and accurate deep neural networks capable of handling unstructured data more practically and commonly. Computer processing power is set to increase drastically in the coming decade, especially with the advent of quantum computing, enabling AI to progress even further. AI-specific microprocessor silicon chips are being developed to enable faster processing of AI algorithms compared to traditional microprocessors. Finally, neuromorphic computing could be also a catalyst in AI progression by introducing brain analogy processing on AI silicon chips.
Like any other industrial revolution, AI will significantly impact our societies, economies, and work environments. The widespread use of AI in most industries will increase global GDP due to organizations becoming more productive and profitable. However, there will be a severe impact on our employment models. Many of the labor-based jobs will be replaced by AI agents, thus leading to mass unemployment and greater global inequality. It is imperative that governments, in conjunction with business organizations, strategize accordingly and mitigate the negative impact of AI in the job markets and societies as a whole. Business organizations will have to emphasize employees’ creativity because AI agents do not do creative tasks very well. Finally, future employees will need re-skilling, and the educational system will need revamping to accommodate digital skills to prepare future workers.
Ethics with AI
We should be concerned about AI agents’ decisions, which can impact society and our well-being. Also, how far will AI reach where we can lose control of it? All this means that AI needs to be regulated. Nation governments, research institutions, and business communities need to implement regulations on the use of AI in our social domains. Data privacy and bias are a concern. Who should access it? How do we prevent data from being misused? How can we make algorithms more democratic in nature? These ethical questions can only be answered and solved by formalizing and implementing ethical-based strategies throughout all communities, industries, and organizations.
Is AI here to stay?
AI, despite the previous AI winters, is here to stay. The availability of data, namely big data, the improved algorithms, the increasing computer processing power and memory storage capabilities have catapulted this technology into a journey of no return.