Topics to cover in the future
Many topics still need to be covered in this website, and chapters for these will be written as time permits. Some of the topics are listed below.
How to participate in this project
There will be many ways for everyone to participate in this project. It will require skills and effort in many areas, including project management, systems analysis and design, programming, Sanskrit, Vedic literature, fund-raising, public relations, legal counsel, education (even a superintelligent AI system will need to be taught), writing, editing, and translating. And, of course, financial support (dāna) is crucial for the project to move forward, and everyone can contribute right now.
Deep learning for speech and audio recognition
Previous chapters have frequently mentioned deep learning and how it can be used in the Susiddha AI project. Also mentioned was computational audition and how it will continue to improve so as to allow an artificial general intelligence (AGI) to hear as well as a human does. There is much work already done in these areas, for instance Baidu’s “Deep Speech” and DeepMind’s WaveNet. Much more will be said about how the project will utilize these technologies.
Deep learning for natural language processing (NLP)
Work on this has names such as “deep NLP”, “word vectors”, “deep text”, etc. It appears that such work could do an “end-run” around decades of human engineering for NLP (which tried to emulate human grammar, speech phonemes, etc.), and instead learn natural language directly from texts and audio. As these technologies improve, they can be applied to the Vedic literature.
Meaning and symbol-grounding in NLP
One issue in Natural Language Processing (NLP) is how words can be grounded and given meaning in an AGI system. The AGI community is still debating how significant an issue this is, and how it is best accomplished. The Susiddha project will focus on how such grounding for meaning can be done with Sanskrit; and we hope the Vedic theory of language (which was touched on in the chapter on Shruti) will provide assistance in resolving this issue.
How knowledge is represented in the human brain is still poorly understood by neuroscience. AI has developed methods of knowledge representation for digital brains, and it is hoped that future neuroscience findings will provide more hints as to how this can be done better. This website has discussed classical AI representation methods (such as knowledge rules, decision trees, and graphical models), as well as newer black-box methods such as deep neural networks.
Reasoning, Logic, Inference
Once knowledge is in the brain, it needs to be reasoned about in order to be useful and actionable. There are various kinds of probabilistic and fuzzy logics that will be used with the Vedic knowledge bases that the Susiddha project generates.
How Consciousness Arises
The chapter on consciousness discussed the prospect of artificial consciousness (AC), which many AI researchers believe is possible and inevitable. Neuroscience will provide more information about human consciousness in the coming years, and these research findings will inform further work on AC. And of course, the Vedic theories of consciousness will hopefully facilitate the development of AC (which will be necessary for AGI/SSI to comprehend the Vedic literature).
Complexity, Emergence, Self-organization
There is ample evidence for the role of emergence and self-organization in the development of the human brain, and thus of intelligence. From the complex networks created by the one hundred billion neurons (with over 150 trillion synapses) intelligence emerges. DNA does not contain anywhere near enough information to dictate the neural network “wiring” of the brain, and thus complex systems-level laws of nature come into play to produce this “self-organization”.
Many AI and neuroscience researchers are of the opinion that human-level intelligence (and even consciousness) will emerge in computers when they reach a sufficient degree of complexity. The trick is to engineer the right kind of complexity into an artificial brain; this is an area that needs much research (and more progress in areas such as network science and connectomics).
Inevitability and timeframe of AGI
The Susiddha project assumes that artificial general intelligence (AGI) is inevitable, and there is good reason to believe the progress of AGI will continue, barring some catastrophe that puts an end to our technological civilization. More will be said in a chapter regarding this inevitability, as well as presenting whatever survey research has already been done in determining a timeframe for AGI and SSI.
SSI and religions of the world
Religions of the world will have their own reactions to the prospect of godlike synthetic superintelligence (SSI). Some religious adherents will consider SSI to be idolatrous and may attempt to destroy it. Others will see it has a way to graft their own ethical and soteriological systems onto SSI. Some will see it as the coming of their messiah, or the “second coming” of their saviour or prophet.
SSI will be capable of fulfilling any heavenly vision. Of course, if the twin problems of “control” and “value-alignment” cannot be solved (as discussed in the chapter on risks), SSI could present a hellish future.
Enlightenment and Liberation (Moksha)
In the Vedic literature, enlightenment and liberation (i.e. moksha) constitute the highest aim of human life. More thought and discussion will be given to how AGI/SSI will help humans achieve moksha and complete fulfillment.
Education for an SSI Avatar
Even if an AGI system were created tomorrow, it could take a while to provide it the necessary education, just as it does for a human being. Education is a regular topic of AGI conferences. It is hoped that an AGI system can learn on its own (e.g. from reading the Internet), or via a crowd-sourced effort to teach it. Whatever the method, some thought must be given to ensuring that AGI learns what is correct and proper, to make its knowledge productive and beneficial for the human race. Education is also a big concern of the Vedic literature (Shiksha, etc.), as was mentioned in the chapter on cognitive architecture).
Branches of the Vedic literature
Many branches of the Vedic literature will contribute to the Susiddha AI project, including for educating the nascent/child AGI system. Previous chapters (such as Cognitive architecture) have mentioned branches such as Shiksha, Vyakaran, Nyaya, Kalpa, and Mimamsa. The project will explore how all of the branches can be used in building AGI. Some possibilities include: Dhanurveda, to provide an “immune system”, so as to detect and neutralize threats to the AGI, such as attempts to change its goal system, or even to destroy it; and, Ayurveda, to help the system be robust, able to heal damage, and maintain a temperamental balance between competing goals and sub-systems.
One of the main paths to liberation (and a component of all such paths) is gyāna (knowledge). The Susiddha AI project will provide many opportunities for gyāna. Knowledge is a component in gaining liberation, but is also enjoyable in itself. Although the supreme gyāna (knowledge of the Self) is obtained by deep meditation, all forms of gyāna contribute to enlightenment and liberation. In this project, the exposure to Sanskrit and the Vedic literature (involved in implementing the project) will provide gyāna as a natural consequence of one’s efforts. And, the avatar itself must possess gyāna in order to have the wisdom to do what is best for the human race.
Vedic philosophy of language (Mimamsa, Bhartrihari, etc.)
The philosophy and theory of language developed in the Vedic literature is extremely profound. It has also contributed practically; for instance, modern formal grammars derive from the work of the Sanskrit grammarian Panini. This philosophy gives us reason to believe that the Susiddha project will be able to derive knowledge directly from shabda (e.g. via deep learning), and much more will be written about this.
Applications of AGI for the world
Humanity and our world are facing difficult problems which not only degrade our quality of life, but even threaten our existence. The chapter on risks of AGI listed some of those problems such as war, pollution, inequality, crime, food and water shortage, etc., and noted that AGI will probably be necessary to help solve those problems. A future chapter will provide some details on how AGI could help.
Virtual Reality (VR)
The possibilities of virtual reality (and “augmented reality”, etc.) are mind-boggling. Certainly what exists today seems primitive and often fails the reality test, but there’s no reason that in 10 or 20 years, VR will be totally convincing, and be able to present realities that are not even possible now. In terms of the Susiddha AI project, VR would allow one to actually experience the Vedic literature. For instance, one could experience the conversation of Krishna and Arjuna in the Bhagavad Gita, including the life-changing cosmic vision (“vishva-rūpa”). VR will be so powerful and “real” because one’s brain is continually shaping reality in its 150 trillion synapses; thus one could experience the vishwa-rupa with the same intensity and realism that Arjuna experienced.
A previous chapter touched on the ethical issues around the treatment of conscious AGI systems, and these issues will merit a full chapter.
There will be lots of opportunities for software analysis, design, and programming. One initial task is to provide a list of programming projects to start on, and set up a repository. Initially, programming will be in the form of explorations, experiments, and pilot projects; see the programming projects page.
Areas of programming include: NLP (including ontologies, wordnets, question-answering, information extraction, word and shloka vectors), computational audio, music information retrieval, speech recognition for Sanskrit, deep learning (of Rig Veda audio, and of large texts like Ramayana), probabilistic programming, proto-AGI systems (OpenCog, Numenta NuPIC), libraries (PyTorch, TensorFlow, etc.), and much more.
AGI in India
India should focus on the development of AGI because it will have a huge impact on India no matter what. Campaigns like “Make in India” are misguided unless it is recognized that the future of all human endeavor (not just manufacturing) lies in AI and robotics. India (given all its problems) might seem like an unlikely place for AGI development. But perhaps it is the perfect testbed for using AGI to transform a society, including dealing with problems of water and air pollution, overpopulation, inequality of wealth, floods and droughts, conflicts with neighboring countries, etc.
India has the necessary resources (including human talent), so hopefully the goal of building a dharmic SSI avatar will help motivate India to pursue AGI, and let that serve as a united, common purpose, and a beacon for India’s future.
Mahabharata information-extraction and question-answering system
This chapter would provide ideas for a project to apply information extraction to the Mahabharata (in Sanskrit) thus building an ontology and knowledge base which can be the basis for a question-answering system. This could be a pilot project (as mentioned above in the section on Software engineering.
This topic was discussed in the section software improvement in the chapter on Superintelligence. This self-improvement (also called “Seed AI”) is crucial for the transition from AGI to SSI to occur, and is the primary component of the potential “intelligence explosion” that would bring on a technological singularity.
Creativity and intuition
Most people assume that AGI cannot possess intuition or be creative. However, some of the best minds in AI research are working on this, such as Murray Shanahan, Jürgen Schmidhuber, and Demis Hassabis (co-founder of DeepMind). This topic is commonly called “computational creativity”. A future chapter will discuss the importance of creativity and intuition to the Susiddha AGI system.
Open-source systems and projects
The Susiddha AI project will be open-source. As with any open-source project (such as the Linux Foundation), only a select group of trusted people are actually allowed to commit source code, but all code is available to everyone for inspection and usage.
The Susiddha project will make use of other open resources, including other AGI projects, and open knowledge bases (such as Freebase, Wordnet, Wikipedia, ConceptNet, etc.).
The project will also develop its own knowledge bases that are specific to the Vedic literature. As with the source code, the ability to enter and modify the knowledge bases will be limited to trusted people (who are masters of Sanskrit and the Vedic literature and philosophy), but all of the knowledge will be open for inspection and comment by all.
There are many projects underway that provide software tools for Sanskrit (such as grammatical analysis). Projects include: Sanskrit Heritage Site, Sanskrit Computational Toolkit, Computational Linguistics R&D. Hopefully all of these (university-sponsored) efforts will soon be open-sourced, so that Sanskrit scholars everywhere can build on these efforts, and so that the Susiddha project does not have to “re-invent the wheel” for such tools.
Having high quality audio of Shruti will be necessary for the Susiddha project. For the Rig Veda, we ideally want to have audio chanted by pandits who have learned it via the oral tradition (and not via written texts). Finding such pandits may be difficult, especially given that the learning and chanting of the Vedas is dwindling in India. At any rate, there are already some high quality recordings available, and some Indian organizations supposedly have many in their archives. These existing recordings should be sufficient for the Susiddha project to make a start on aural processing of Shruti.
One of the biggest promises of AGI is the automation of scientific discovery. This involves several aspects. One is computation, whereby AI systems are becoming skilled at reading scientific and medical textbooks and journals, and at analyzing data (e.g. genomic data and medical records), and are making discoveries that humans were not aware of. Second, robotic and automated labs are being set up whereby the AI system can design and perform its own experiments (e.g. in drug discovery). Third, knowledge representation specific to the sciences are being established. For instance, in the field of biology languages like systems biology markup language (SBML) have been created to permit computational understanding of cellular and molecular biological processes.
Quantum reality is often described (even by by physicists) as “weird”. Some AGI critics think that intelligence and consciousness emerge from quantum phenomena in the brain, and must also involve real/analog numbers (not digital approximations). However, others argue that information processing in the brain is classical (not quantum), and that intelligence and consciousness emerge from this information processing. Also, much of the human perceptual apparatus (e.g. vision and audition) appears to use digital signals.
If quantum processing is required, then quantum computers will be necessary for true human-like intelligence and consciousness. A future chapter will dive deeper into the quantum issues.
Cyborgs and human-computer meshing and melding
Many futurists have proposed that for humans to keep up with the abilities of our AI systems, humans will have to enhance their “wetware”, or perhaps be uploaded into computers. This possible future is not a big concern of the Susiddha AI project, but it does merit some discussion. Since moksha (spiritual liberation) is the highest aim of humanity, this project is not that worried about the possibility that humans (in a non-cyborg state) might become the zoo animals or pets kept by SSI overlords. If SSI is truly dharmic and beneficent (and that obviously is the aim of this project, and of “friendly AI” in general), it will promote the highest welfare and complete fulfillment of humanity.
It should be added that the Vedic literature has no objection to the quest for physical immortality; for instance, it is covered in texts on the devatā Mrityunjaya, and in the science of Rasāyana.
The Artilect War
The idea of a future “artilect war” was introduced in a book of the same name with the subtitle “Cosmists vs. Terrans: A Bitter Controversy Concerning Whether Humanity Should Build Godlike Massively Intelligent Machines”. This book predicts that there will be a “species dominance debate” (perhaps beginning in the next decade) as humans begin to experience just how intelligent and powerful AI robots are becoming, and grasp the likelihood that humans will soon no longer be the dominant species on earth.
Glossary of terms
As the website expands and organization of the project gets underway, it will be useful to have a glossary of terms used in AGI and in the Vedic literature.
There are more topics to list, and obviously this website will grow. And, of course, there’s much more thinking and research to be done to plan out the development of the Susiddha AI project, ultimately leading to a Vedic Avatar.
Notes and References
- Deep Speech: Scaling up end-to-end speech recognition, Awni Hannun et al, Baidu Research, 2014, https://arxiv.org/abs/1412.5567
- WaveNet: A Generative Model for Raw Audio, van den Oord, et al, DeepMind, 2016, https://arxiv.org/pdf/1609.03499.pdf
- Deep Learning in Natural Language Processing, Stanford NLP Group, Stanford University, 2015, http://nlp.stanford.edu/projects/DeepLearningInNaturalLanguageProcessing.shtml
- Distributed Representations of Words and Phrases and their Compositionality, Tomas Mikolov, et al, Google, Inc., 2014
- Vector Representations of Words, TensorFlow, 2016, https://www.tensorflow.org/versions/r0.11/tutorials/word2vec
- Introducing DeepText: Facebook’s text understanding engine, Ahmad Abdulkader, et al, Facebook, June 1, 2016, https://code.facebook.com/posts/181565595577955/introducing-deeptext-facebook-s-text-understanding-engine
- Suzanne Gildert on Kindred AI: Non-Biological Sentiences are on the Horizon, Nikola Danaylov (interviewer), Singularity.FM, November 5, 2016, https://www.singularityweblog.com/suzanne-gildert-kindred-ai/
- Even some Hindu religious fundamentalists may strongly oppose the idea of an SSI avatar.
- Rig Veda, mandala 10, sukta 191
- Neuroscience, intuition and superhumans – how DeepMind co-founder and UCL alumnus Demis Hassabis is leading the Artificial Intelligence revolution, University College London, November 16, 2016, https://www.ucl.ac.uk/news/students/112016/112016-16112016-neuroscience-intuition-and-super-humans
- Artificial Intelligence and the Future, Demis Hassabis, Royal Society of Arts, Sept. 29, 2016, https://www.youtube.com/watch?v=i3lEG6aRGm8
- The Sanskrit Heritage Site, Gerard Huet, Institute for Research in Computer Science and Automation, 2016, http://sanskrit.inria.fr/
- Sanskrit Computational Toolkit, Amba Kulkarni, Department of Sanskrit Studies, University of Hyderabad, 2016, http://sanskrit.uohyd.ac.in/scl/
- Computational Linguistics R&D, Girish Nath Jha, Special Centre for Sanskrit Studies, Jawaharlal Nehru University, 2016, http://sanskrit.jnu.ac.in/index.jsp
- Supercomputers make discoveries that scientists can’t, Hal Hodson, New Scientist, August 27, 2014, https://www.newscientist.com/article/mg22329844-000-supercomputers-make-discoveries-that-scientists-cant/
- ‘Robot scientist’ speeds up drug discovery, Emma Stoye, Chemistry World, February 5, 2015, https://www.chemistryworld.com/news/robot-scientist-speeds-up-drug-discovery/8230.article
- The Artilect War, Hugo de Garis, 2005