Understanding AI And Its Impact On Industries
Introduction to Artificial Intelligence
Artificial intelligence (here in after referred as AI) is an old knowledge set, but numerous imperative and innovative technologies are emerging which are the results of compute, big data, and cloud storage (Cristianini, 2016). AI technology is a liberatory by core nature and industries those assimilate it, will found their workers more advanced, creative, and greatly adaptive than ever before. However, these above technologies are still in initial stages and there is a long way to achieve ahead. In this report, we will examine important technological issues as well as social issues related to AI for all main industrial areas while integrating AI into apps, medical procedures, automated innovation, business intelligence, daily leisure ness, and key business processes in order to support human decision-making (Bengio, 2016).
Although, the aim of AI implementation is society beneficial motivational study in all areas, from security and control to economics, law, technical topics, security and control. However, Short term risks attached with AI can be explained with this example: If your laptop is controlling your airplane, car, pacemaker, automated trading structure and the power grid then it will definitely give awesome results. On the other hand, it may result in a major nuisance if the laptop either crashes or gets hacked and all technological benefits through AI will be vanished at that moment. Similarly there are so many long term risks attached with AI, which will be discussed in the below sections of this report (Clickatell, 2017).
To understand AI as a Machine Learning (here in after referred as ML), first we need to understand AI and ML separately then only we can relate the two (Jones, 2018). AI is defined as the capability of a machine to execute cognitive functions of human brains such as learning, reasoning, perceiving and problem solving. Some key technologies that allow AI to resolve business issues are as follows:
- Machine Learning
- Robotics and self-governing vehicles
- Virtual agents
- Computer vision (McKinsey Analytics, 2018)
ML algorithms identify patterns and utilize those patterns to learn predicting and recommending by data processing and best experiences rather than following external programming instructions. The interesting fact is that these algorithms also get adapted easily to new information and experiences in order to improve efficacy (Reynolds & Day, 2018). Therefore almost all recent advancements in AI have been accomplished through applying ML theory to very huge data-sets. ML is a subclass of AI, this can be further explained as all ML can be counted as AI, but not all AI can be counted as ML (Pyle & San Jose, 2015).
Short-term and Long-term Risks of AI
ML uses prior learning and provides predictions as well as prescriptions through a number of analytics such as Descriptive, Predictive and Prescriptive.
Major types of ML are as follows:
In this type of ML, algorithms use training data set and feedbacks from humans in order to learn the relation in between given inputs and the output. Whenever you need predictions and behavioural understandings from new data then this algorithm calculate through supervised learning technique.
The mechanism of this kind of ML can be explained as:
Step-I: Human beings indicate each element of the input data set and also describe the output variables.
Step-II: Training of this algorithm is done on the basis of above data sets in order to identify the relationship between input and output variables.
Step-III: After completion of training, and testing of algorithm accuracy, it is applied to a new data set (Schölkopf, 2015).
This ML algorithm usually explores input data sets without giving am external output data set. Classification of data, identification of patterns can be obtained through this algorithm. The mechanism of unsupervised learning can be explained as follows:
Step-I: the algorithm accepts unlabelled data and utilizes these data sets to structure image of data.
Step-II: after analysing the unlabelled data sets, it gives a structure from the input data.
Step-III: the algorithm examines a group of data those exhibit same behavioural characteristics (Jones, 2018).
This ML algorithm learns through received rewards on its actions. It will perform a task in a way, in which, when it had performed earlier and got positive results. Whenever you want to explore an area and at the same time you do not have training data sets and still you are not able to portray the ideal final state, then you may use this algorithm.
Working principle can be further explained in following steps:
Step-I: Initially, algorithm attempts an action on the environment around it.
Step-II: after attempting, if this action is in the positive direction then reward will be added and increase previous rewards available.
Step-III: finally, the algorithm will optimize the best series of events through re-correcting itself over time (McKinsey Analytics, 2018).
Definition of deep learning is that “it is a kind of ML which can process a broader range of data set sources and can generate more accurate results than conventional ML methodologies.” Deep learning does not require data pre-processing by human beings. Neuron which means interconnecting layers of soft wares based calculators here, form neural network of deep learning ML (Gopnik, 2017).
Supervised and Unsupervised Learning
Deep learning uses following steps to execute process and give results:
Firstly, neural network ingest wider input data sets and process those from multi layers. During multi layers algorithm processing it explores the data deeply and learns complex features of the input data sets at every layer (Wong, 2016).
During second step, this network makes a structure of the provided data and then learns about its accuracy, and analyse learning during thorough process.
- Convolutional Neural Network (referred as CNN)
- Recurrent Neural Network (referred as RNN)
- Feed forward Neural Network (referred as FFNN) (McKinsey Global Institute, 2018)
The deep learning AI techniques are the techniques which are based on ANN (Artificial Neural Network). These techniques are generating 40 per cent of the whole potential value that can be provided by all available analytics techniques. The worth of AI cannot be calculated in the models of the AI, but it lies in organizations’ capabilities to join them. Hence, professional leaders will require arranging and selecting choices carefully about deployment of them. The data usage must always be done with concerning on the following issues:
- Data security issues
- Privacy issues
- Potential issues of bias
DL Techniques those address estimation, classification, and data collection issues are presently the most extensively applicable in the use cases as we recognized, reflecting the glitches whose resolutions drive value across various sectors (McKinsey Global Institute, 2018).
The highest potential for AI Deep learning technique is to generate value in use cases. In these use cases already established analytical techniques like classification and regression techniques can be used, but these ANN techniques can provide more enactment and generate surplus insights and uses. According to the research data it is evident that, 69 per cent of the AI use cases identified , out of which only 16 per cent of these use cases we found as a greenfield AI result that were highly appropriate where other analytical methodologies would not be operative and effective.
To capture the value impact of these DL techniques, we require multiple problems solution. Technical limits are including a large volume and multiplicity of labelled training dataset requirement, although presently a lot of efforts putting are helping address these. Societal issues and regulation, for an example while using personal data, data security of personal data is a big constrain in AI use within insurance, banking, health care, medical products and pharmaceutical as well as in the public and social sectors, if the above issues are not correctly solved. The ruler of the value economic and societal influence creates a command for all the contributors such as AI innovators, AI-assessing companies and AI-policy-makers to certify a lively AI atmosphere which can safely and effectively capture the financial and social welfares (McKinsey Global Institute, 2018).
Convolutional Neural Network
Machine Learning and its Types
A CNN is a multi-layered neural network with a distinct design in order to abstract progressively complex structures of the data sets at every layer to define the correct output.
Utility of CNN is high, where you have an amorphous data set and you require inferring efficient information from it (McKinsey Analytics, 2018).
Processing an image through CNN
- The CNN collects an image for an example, of a laptop and that it practises as an assembly of pixels.
- In the inner hidden layers of the CNN model, it recognizes exclusive features, such as the structure and outline image of that laptop.
- The CNN will now categorise a different image as of the laptop if it will find unique features in the prior shown image to it (McKinsey Analytics, 2018).
Diagnose health problems from medical image scanning
In this business model of CNN, deep learning is introduced from a radiology outlook. However, when address the utilization of AI in medical imaging; we expect that the CNN technological innovation will serve as a cooperative medium by lessening the problem and disturbance from several repetitive and monotonous tasks, rather than just substituting radiologists (Lee, et al., 2017).
The use of deep learning CNN and Artificial intelligence in radiology is presently in the phases of infancy. With the current technological innovations through Image Net, huge and entirely annotated data sets are desirable for evolving AI development in medical scans. This will be important for training the CNN, and also for its assessment. The energetic participation of efficient radiologists is also needed for founding a great medical scanning datasets. Additionally, there are countless other issues and practical difficulties to resolve and overcome. Therefore, legal, ethical, and regulatory problems raised in the usage of patient medical scanning data for the progress of AI deep learning should be wisely considered. This business model of CNN is very innovative and has wide scope of improvement and innovation as well as discovery as per the viewpoints of several radiologists, scientists, law and ethics principles experts and engineers, (Lee, et al., 2017).
Detect defect and inspect products in a steel production line through Real-time image processing
This business model introduce AI deep learning through real-time image processing in order to inspect edges and detect defect in stainless steel production lines (Spinola, et al., 2011). Deep learning CNN can use an image scanning and handling system to calculate the width and examine the quality class of the stainless steel stripe in a production sector for reducing human efforts, time and enhancing quality (Dickson, 2017). Real-time image processing of the image scanned attained through a twin camera system will generate image and analyse. Image processing algorithms based on CNN detect defective products through edge inspection. This system will be quality enhancement and quality control innovation in a stainless steel production line (Spinola, et al., 2011).
Conclusion
The discussion is heading to the conclusion that there are numerous advantages and dark sides of AI enabled technologies. Desirable is that we will recognize the great challenges that lay in front of us and confess our duty to ensure that we will take whole advantage of the innovations while decreasing the trade-offs (MIT Technology Review, 2017). On the other hand, it can be sensed in a way that the machines are coming in a form of robots, but we will not let them rule over human society. We will use this technological aspect of AI in such an extent that it will be executing aiming to bring peace worldwide. While machines will reduce human efforts, they will also bring disruptive modifications and will raise new complications that can affect the economical, ecological, legal, moral and ethical scenario of human societies.
Companies and sectors, which are utilizing AI, enabled technologies at huge level need to address these following areas critically for future: Jobs and employment, biasing issue, responsibility, security and privacy.
References
Bengio, Y., 2016. MACHINES WHO LEARN. Scientific American, 314(6), pp. 46-51.
Clickatell, 2017. Trends in artificial intelligence technology. [Online]
Available at: https://www.clickatell.com/articles/technology/trends-artificial-intelligence-technology/
[Accessed 25 09 2018].
Cristianini, N., 2016. A different way of thinking. New Scientist, 232(3101), pp. 39-43.
Dickson, B., 2017. 4 challenges Artificial Intelligence must address. [Online]
Available at: https://thenextweb.com/artificial-intelligence/2017/02/27/4-challenges-artificial-intelligence-address/
[Accessed 25 09 2018].
Gopnik, A., 2017. Making AI more human. Scientific American, 316(6), pp. 60-65.
Jones, L., 2018. Artificial intelligence, machine learning and the evolution of healthcare. Bone & Joint Research, 7(3), pp. 223-225.
Lee, J. et al., 2017. Deep learning in medical imaging: general overview. Korean journal of radiology. Korean journal of radiology, 18(4), pp. 570-584.
McKinsey Analytics, 2018. An executive’s guide to AI. London: Mc Kinsey & Company.
MIT Technology Review, 2017. The AI Issue. [Online]
Available at: https://www.technologyreview.com/s/609123/the-ai-issue/
[Accessed 25 09 2018].
Pyle, D. & San Jose, C., 2015. An executive’s guide to machine learning. 3 ed. London: Mckinsey Quarterly.
Reynolds, R. & Day, S., 2018. The growing role of machine learning and artificial intelligence in developmental medicine. Developmental Medicine & Child Neurology, 60(9), p. 858–859.
Schölkopf, B., 2015. Learning to see and act. Nature, 518(7540), pp. 486-487.
Spinola, C. et al., 2011. Real-time image processing for edge inspection and defect detection in stainless steel production lines. Imaging Systems and Techniques(IST),2011 IEEE International Conference, pp. 170-175.
Wong, W., 2016. A deeper look at deep-learning frameworks: in artificial intelligence, deep learning continues to gain ground, thanks to multicore hardware such as GPGPUs, with tools and frameworks also providing more accessibility to the technology. Electronic Design, 64(8), p. 28.