For example, Linear Regression is used for Machine Learning to predict values that will occur in the future given an input, or Logistic Regression is used to classify complicated data or images of malignant tumors. In the future, it is expected that AI will be able to do programming or self-improve with its own codes. The experiments adjust or change the connection between sensory organs and the corresponding parts of the brain.
With this architecture, data scientists create an algorithm that mimics the way the brain works and becomes the neural network algorithm as shown in the picture. In the function, there are weight and bias variables and they will be useful for adjusting the function to fit a particular data set.
Unsupervised Learning
Neural network uses a technique called Backpropagation to help optimize the cost function or in other words, adjust the weight and bias variables. Backpropagation works backwards by changing the neural net through the weight and bias variables in each node of each layer so that the output is as close as possible to the training data. After retrieving the centroids of all the centroids, each will be moved to the corresponding centers, thus the centroids are moved to the center of their elements.
Reinforcement Learning or Deep Reinforcement Learning
Next, the center of gravity for each centroid and its elements is calculated by finding the average of all the distances. This process repeats itself until centroids cannot be moved or slightly from the previous coordination. We can use concepts of exploration/exploitation trade-off, Markov decision processes (MDPs), Q-learning and Value learning problem for machine agents to be able to do assigned tasks.
Epsilon is the percentage that the agent will randomly perform the chosen actions instead of those to maximize the reward. The larger amount the agent learns corresponds to the smaller epsilon value we set, so it optimizes the result. The set of all possible states of an agent, such as the position of a rat in a maze, a dead end, and a crossroads.
A set of actions available in each state, such as turning left, right, U-turn and going straight. For example, the agent turns right at an intersection and ends up in a new position. Learning rate alpha corresponds to linear regression, which is how aggressive the update we want it to be.
Deep reinforcement learning is actually reinforcement learning, but the nodes in the hidden layer are much more in number because they are required for the algorithm to solve complex problems.
Recurrent Neural Networks
The picture above shows how the recurrent neural network is implemented, X is the input, O is the output, W is the weight and S is the states. You can see that within a neuron, there are three sub-neurons, which contain values of various other neurons, and all are calculated from the weight of the main neuron. Considers given letters by placing them in S state and predicts gap possibilities.
Genetic Algorithm
RESEARCH METHODOLOGY
The Source and Sample Size
Instrument
All the information derived from the AI self-encoding group will be summarized and constructed in a comprehensive way for the reader of this thematic paper. Details will also be placed in the right places to complete the process of creating a self-modifying AI. The Business Impact group information will be summarized highlighting the most targeted business or sectors and the reasons for the impact.
How will self-modifying AI impact and change the way people do business or work in Thailand?
RESEARCH FINDINGS
Interviewee 4
AutoML tries and tests different algorithms to find the best fit for the problem and then refine them further, which takes less time and effort than the same work done by machine learning experts. Speaking of the time and effort required to create an AI model, typically a deep learning model, a 10-layer network can have about 10 to the power of 10 versions, and this can take years for teams of the best field researchers and engineers to calibrate. The child model will be tested so many thousands of times in order to have the optimal result.
The architecture of the AI model for kids is similar to the architecture created by top machine learning experts. Talking about the effects of self-coding AI on businesses, machine learning services will be available for people who do not have much knowledge about machine learning and have money to hire machine learning experts. Perhaps software developers will be the jobs affected if programs can literally create programs.
First, the role of developers will change because it will be easier to create a program to solve business problems. Automation related to machine learning will be ubiquitous and developers will have to learn more about the Internet of Things (IOT), that is, the information we can receive from various devices. In addition, today's complex software will be available to more people in the future, so small to medium and medium to large businesses will have greater access to software.
In addition, education will gradually change as more people gain access to the Internet, and there will be applications that support students in learning.
Interviewee 5
CONCLUSIONS
Conclusion
- The make of self-coding AIs
- Business Impact
Therefore, reinforcement learning with neural networks is inevitably chosen to create AI that can create a program or AI. One disadvantage of this approach may be the fact that we may not know how the AI arrives at an answer or result, but we can use it. The first approach is to use a genetic algorithm to create a program that can write codes from scratch, and the second approach is to use a deep reinforcement learning algorithm to do so.
Deep reinforcement learning algorithm used to create AI that can write an AI model that consists of a few algorithms, which are neural networks, deep neural networks, and recurrent neural networks. The underlying algorithms that use deep reinforcement learning are neural networks, which are similar to how animal brains work. The effect of recurrent neural network is that the AI can calculate current actions by considering various results of various other current, past and future actions.
According to interviewee 1, it may take ten years or more before the AI is actually applied. The AI's ability is the same as that of the developers, but with more accuracy and speed. So as AI learns to code with more programming languages, more developers will have to adapt and change their roles.
This does not exactly mean that developers will lose their jobs, but they must change from the coding task to controlling or inspecting the AI.
Recommendations
As a consequence, developers have no choice but to adapt and in the long run the demand for developers will decrease. However, the demand for people who have skills in machine learning and AI application will increase corresponding to the high rate of AI adoption by companies.
Limitations and Suggestions to Future Research
In my opinion, it is likely that in the future there will be a breakthrough in hardware technology that will allow machine learning to leverage more of the hardware performance. Many countries and major companies have also invested heavily in machine learning, which is accelerating research into machine learning algorithms. Therefore, we could be closer to the application of AI that can write codes with any programming language with new concepts and algorithms, or there could be a research focusing on AI that can help data scientists decipher new algorithms to solve problems better.
Moreover, in practice, Thai businesses will have to respond and use technology and deal with the effects. During that time, we would see more Thai companies applying AI and more data scientists in Thailand. Collecting data for research would be easier, thus producing better and more accurate analysis and results.
As a result, more insights and concepts of self-encoding AI and more detailed impact on business sectors can be explored even further.
Deep Learning: With massive amounts of computing power, machines can now recognize objects and translate speech in real time.
APPENDICES
Interview Scripts
- Background Work
- Background Work
- Background Work
- Self-coding AI
- Background work
As far as I know, this research will mainly focus on the algorithm itself, but it's hard to find one in Thailand because most people I know are doing projects that only use the algorithms created by foreign data scientists. Application developers who do not have advanced knowledge will have to adapt and learn more skills because it will be a lot easier for business people to use software. In the future, we humans will depend a lot on our creativity and social skills to compete in the market.
Then the data set will be used for testing and the test result will be sent to the AI creator to optimize the created AI. The child AI will be tested and modified by the controller AI until it gives the expected result. So I think the role of developers will change and the importance of the work will be hindered.
In the future, data collection, like internet of thing, will be another field that creates jobs, so there will also be vacancies and people need to learn new skills. Education is also an area that will be affected by the technology, especially AIs and increase in internet access. Students can learn things they are interested in without much help from teachers because they can find the resources on the internet and there will be many learning support systems available in the future such as Multi Online Open Courses (MOOC).
The sector that will be most affected is IT sector because if there are AIs that can create codes and also AIs, then many of the IT jobs will be replaced, such as programmers and system administrators.
INTERVIEW CONSENT FORM
