Recent times, we have seen how IoT devices and AI-based technology have been booming. Everyone is in awe to see its applications in multiple domains ranging from Medicine, finance, Vehicles to AI-based security. But the spotlight is on Medicine, and there few areas where its been completely looked down on and given less importance. I have picked up these areas and have researched and studied their progress in the same. The areas are Security & Privacy and Quantum computation and information science.
These areas are the least funded and most overlooked in the present times. But in the near future, these will be the most talked about. With the growth of AI, security is a big concern, the privacy of users must be maintained, and the algorithms must be foolproof and should be secured from any breaches. The work of security in AI and vice versa is slowly gaining momentum while the Quantum realm has still a long way to go, I have summarized both the domains and their progress in AI through a literature survey.
Under the umbrella of AI comes many disciplines such as Machine Learning, Deep Learning and Neural Networks, Natural Language processing etc. With the present trend, AI has been promising to solve and help solve many problems and has a wide area of applications in the fields of Medicine, robotics, vehicles, security etc. A day does not go without any groundbreaking achievements in these fields. For the past few and coming few years, the area seems promising and will remain in the trend.
Security & Privacy
Deep Neural Networks are the core of Computer Vision and Pattern Recognition and help solve the most challenging classification tasks. However, the algorithms are fooled by perturbing them with fake images or adversarial images or videos that could lead the algorithms to end up with wrong conclusions. An adversarial image is a modified version of the original that is disturbed, for example, by adding noise to confuse the system. This is also called the black box attack. The attacker has no clue about the model or algorithm whatsoever and tries to target it by putting in these adversaries. Also, recent advancements of TensorFlow or Keras has made tampering with images and videos relatively easy. We can find DeepFake videos over the internet, which look so like the original that we cannot say the difference. For example, a short video of a former US president went viral for saying things that he has never said. Hackers and curious people always try to break the system and play through the vulnerabilities that exist. Some hackers also use these technologies to get information about people and their data through phishing or other means. Thus security will become a primary concern in the future with this development seen in AI. A decent study has to be made to see how security can be developed using AI and how to secure ourselves and data from AI-based machines and algorithms. As a matter of fact, defence is one of the most funded sectors in the economy, and it will remain.
AI’s future does not hold only for present case technologies of classic computers. Still, many machine learning algorithms and technology could be a step to build or learn things that we do not have sufficient knowledge to implement or understand. The quantum domain is slowly progressing and will be in the talking in the coming future. Quantum theory leads to remarkable and counter-intuitive phenomena in quantum information science and quantum computation. In various cases, it has been significant to reduce problem complexities and perform high computational tasks. Machine Learning and AI tasks can rely on such problem-solving strategies to lead to an advantage. Quantum computers offer many optimization tasks and make searching very easy, like Grover’s algorithm, storing large chunks of data, simulating quantum systems, etc. Properties of entanglement make it unique as there is no accurate correlation in classical computers. It leads to improving security and secure communication using Quantum teleportation. Recent developments in Quantum Machine learning show the building of a state vector machine for big data classification on a quantum computer.
Roman V. Yampolskiy et al. (2016) predicted analysis on the future of AI-based on a timeline of AI failures. Depending on the level of criticality, the system must be handled as they have concluded that there is no system with a cent percentage of safety. A system will eventually fail, but the goal is to reduce the impact and reduce the number of such attacks.
Anagha Kulkarni et al. (2018) describe how to secure the internal data from outside obfuscation in an adversarial environment. The paper talks about minimizing the leakage of information while completing or reaching the goal or specific set of actions.
Ian Goodfellow et al. (2017) talk about how malicious adversarial breaches can force the wrong classification of data and their effect on neural network policies. Since the input set is not fixed, even the smallest, even the smallest, initializing weights affects training, unlike supervised learning. Hence to avoid the situation of a breach during input, some adversarial examples are added.
Deepfake videos can be found all over the internet, and it resembles the original so much that it is challenging to say it apart from the original. Yuezun Li et al. (2018) have come up with a deep learning-based method that can tell them apart.
Another method to detect them has been stated by David Guera and Edward J. Delp(2018) using recurrent Neural networks. Usually, autoencoders are used to come up with deep fakes. Two sets of images can have a face swap by encoding it and using the decoder of set two on one to swap the latent factors that make a face swap.
Haya R. Hasan et al. (2019) tried to combat deepfakes using blockchains and smart contracts. Due to the authenticity of blockchain and smart contracts, they being completely decentralized, they can be used to identify an original from a fake. Smart contracts try to trace back to the origins of the file and track the digital footprint of it across its timeline to point out its authenticity. The metadata is stored across the chain, and credibility can be assessed if it comes from a trusted source or not.
Valentina Zant et al. (2017) describe efficient strategies against adversarial attacks based on some practical observations that perform better than a state of the art defences and is easily implementable.
Jonathan Blackledge et al. (2015) have come up with algorithms of neural networks that help to encrypt data. Using artificial neural networks for cipher generation is non-linear and is quite analogous to produces pseudo-random numbers, and the weights here become the keys.
Fangzhou Liao et al. (2018) have proposed a High-level representation guided denoiser as a defence for classification. They have come up with a loss function that comes up with outputs activated by clean and denoised images. Usually, adversaries are inserted in during the input. It is good to denoise them before they are sent in for an end-to-end process to obtain the output.
With a lot of our lives being spent on intelligent devices, the question to privacy and security of personal data arises, and Xiaokui Shu et al. (2015) have come up with algorithms that can help detect such leakage of personal data and security to our privacy being maintained. Since most data leaks happen due to human errors, they have devised a Network Data leak detection by performing deep packet inspection and searching for sensitive data patterns.
Xiaokui Shu et al. (2015) attempt to quickly detect any data leaks. They consider data transformations where the data could have been made to seem different from the sensitive data by observing data patterns based on the scale of data transferred and the amount of content.
Vedran Dunjko et al. (2018) talks about how the domains of Quantum information and Artificial Intelligence can advantage of each other to grow in the future. Quantum computing enriches the speed of the algorithms and the amount of memory it can work with. With the expanding Big Data analysis, the state of art technologies can benefit loads if they can be merged.
Quantum Aspects of Semantic Analysis and Symbolic Artificial Intelligence tries to connect aspects of artificial mathematical representations and scientific disciplines of semantic analysis and AI to quantum information technology. It talks about the similarities of the SVD(Singular valued decomposition) a technique where for example is used to break down data into multiple fragments such as users and ratings into matrices a form of eigen valued decompostion where users with less eigen value can be emitted for easier manipulation , QIT algorithms and protocols have a similar representation like vectors and can be used to carry out such algorithms on a very large corpus. Also another technique similar to SVD is the Latent Semantic Analysis which is popular technique in the world of information retrieval and can be strongly associated with the Hilbert problem in the world of quantum computation.
Quantum computation, quantum theory and AI  gives a brief overview of the different aspects of the three mentioned amidst each other. For unavoidable exponential increasing decision trees, the Hamiltonian evolution tries to figure out solutions way more efficient than classical ones. Grover’s algorithm helps in searching target states or searching in general, viewing the information as quantum superposition and using the concurrency of a quantum domain, we can solve many problems to our advantage. Due to the probabilistic nature of qubits, statistical inference is at the heart of it. The paper clearly describes the breakdown of the quantum domain that can be used to efficiently solve AI problems and some possible areas that can be explored, which could be used.
Yu-Bo Sheng et al.(2017) gives a security perspective in quantum machine learning and how concepts of quantum teleportation, quantum key distribution etc. can be used. As we expand the power of the quantum processor, high dimension vectors can be sent, which eventually is our motive for the future. The protocol takes advantages of the features of entanglement, encoding and decoding using polarization, application of applying quantum gates such as CNOT gates.
Vedran Dunjko et al.(2016) have devised algorithms for Quantum-enhanced Machine Learning techniques in clustering problems and classification problems. Also, in Reinforcement models where the search space is viewed as Hilbert space, the set of states form a Hilbert space and have observed how the agent and transitions are seen in the quantum domain.
With so many examples where quantum computers help make machine learning better, Alexander Hentschel and Barry C. Sanders(2010) have devised an algorithm to use Machine learning for quantum measurement.
Most of the funding even in a country is done for defence in general thus, when you think for the future of AI, I have assumed that most of the research interest should lie in Cybersecurity, security for and by AI and improving the budding technology of quantum computation and information science to view the world in a non-classical way.
Attackers always try to meddle with the system to suit their own needs, and building a complicated system makes it even more difficult as even the smallest of the attacks can have significant impacts. Thus implementing a foolproof and decentralized system should aim for the future, and integrating the learning of various fields must be looked upon. Security should be implemented from the very beginning stage, and the system should not be entirely susceptible to attacks and breaches. We saw how generating adversaries could help the algorithm learn better of the original and fakes. Smart contracts for identifying social footprints and maintaining privacy and log of data being sent and received is essential. Though there is no way for 0% error or a 100% success rate, the driving force must minimize the former and maximize the latter with all the tech giants claiming quantum supremacy.
When the world is ready with quantum computers, the view and scene of everything are going to change. Quantum computers will not wholly replace smart devices or classical computers but will significantly impact computational power, security and memory. Experimenting and working on small scale applications will eventually benefit the future yet to come. It does not mean we completely shift the focus towards the quantum domain but seeing the growth of Artificial intelligence in the present era. We can adapt to the changes quickly and efficiently.
 Kulkarni, A., Klenk, M., Rane, S., & Soroush, H. (2018, February). Resource bounded secure goal obfuscation. In Proceedings of the AAAI Fall Symposium on Integrating Planning, Diagnosis and Causal Reasoning, New Orleans, LA, USA (pp. 2–7).
 Cheng, Z., & Hurley, N. (2009, October). Effective diverse and obfuscated attacks on model-based recommender systems. In Proceedings of the third ACM conference on Recommender systems (pp. 141–148).
 Zantedeschi, V., Nicolae, M. I., & Rawat, A. (2017, November). Efficient defenses against adversarial attacks. In Proceedings of the 10th ACM Workshop on Artificial Intelligence and Security (pp. 39–49).
 Akhtar, N., Liu, J., & Mian, A. (2018). Defense against universal adversarial perturbations. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 3389–3398).
 Yampolskiy, R. V., & Spellchecker, M. S. (2016). Artificial intelligence safety and cybersecurity: A timeline of AI failures. arXiv preprint arXiv:1610.07997.
 Blackledge, J., Bezobrazov, S., & Tobin, P. (2015, July). Cryptography using artificial intelligence. In 2015 International Joint Conference on Neural Networks (IJCNN) (pp. 1–6). IEEE..
 El-Zoghabi, A., Yassin, A. H., & Hussien, H. H. (2013). Survey report on cryptography based on neural network. International Journal of Emerging Technology and Advanced Engineering, 3(12), 456–462.
 Huang, S., Papernot, N., Goodfellow, I., Duan, Y., & Abbeel, P. (2017). Adversarial attacks on neural network policies. arXiv preprint arXiv:1702.02284.
 Brendel, W., Rauber, J., & Bethge, M. (2017). Decision-based adversarial attacks: Reliable attacks against black-box machine learning models. arXiv preprint arXiv:1712.04248.
 Liao, F., Liang, M., Dong, Y., Pang, T., Hu, X., & Zhu, J. (2018). Defense against adversarial attacks using high-level representation guided denoiser. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 1778–1787).
 Sameen, M., Han, K., & Hwang, S. O. (2020). PhishHaven — An Efficient Real-Time AI Phishing URLs Detection System. IEEE Access, 8, 83425–83443.
 Shu, X., Yao, D., & Bertino, E. (2015). Privacy-preserving detection of sensitive data exposure. IEEE transactions on information forensics and security, 10(5), 1092–1103.
 Shu, X., Zhang, J., Yao, D. D., & Feng, W. C. (2015). Fast detection of transformed data leaks. IEEE Transactions on Information Forensics and Security, 11(3), 528–542.
 Li, Y., & Lyu, S. (2018). Exposing deepfake videos by detecting face warping artifacts. arXiv preprint arXiv:1811.00656.
 Hasan, H. R., & Salah, K. (2019). Combating deepfake videos using blockchain and smart contracts. Ieee Access, 7, 41596–41606.
 Güera, D., & Delp, E. J. (2018, November). Deepfake video detection using recurrent neural networks. In 2018 15th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS) (pp. 1–6). IEEE.
 Dunjko, V., & Briegel, H. J. (2018). Machine learning & artificial intelligence in the quantum domain: a review of recent progress. Reports on Progress in Physics, 81(7), 074001.
 Aerts, D., & Czachor, M. (2004). Quantum aspects of semantic analysis and symbolic artificial intelligence. Journal of Physics A: Mathematical and General, 37(12), L123.
 Venegas-Andraca, S. E., & Bose, S. (2003, August). Quantum computation and image processing: new trends in artificial intelligence. In IJCAI (p. 1563).
 Dunjko, V., Taylor, J. M., & Briegel, H. J. (2016). Quantum-enhanced machine learning. Physical review letters, 117(13), 130501.
 Ying, M. (2010). Quantum computation, quantum theory and AI. Artificial Intelligence, 174(2), 162–176.
 Sheng, Y. B., & Zhou, L. (2017). Distributed secure quantum machine learning. Science Bulletin, 62(14), 1025–1029.
 Hentschel, A., & Sanders, B. C. (2010). Machine learning for precise quantum measurement. Physical review letters, 104(6), 063603.