The danger in stating which AI method will be the leader in next years is to think that every problem is a nail and we just have a hammer. This has been the problem with the current emphasis on deep learning.

Each problem is unique and must be solved with the appropriate approach, methodology and tools that are adequate for the problem (just take as example the no free lunch theorem in ML). AI is not ML and includes a multiplicity of algorithms and heuristics that in some instances cannot be compared side by side.

although there are many developments in AI, I personally think that there is still a lot to be done in this area. in other words, it is just a beginning for a new world!

Dear Sajad Saraygord Afshari .

Look the link, maybe useful.

https://www.forbes.com/sites/bernardmarr/2018/12/03/5-important-artificial-intelligence-predictions-for-2019-everyone-should-read/#6ab199d1319f

https://hackernoon.com/top-trends-in-2018-you-need-to-follow-for-artificial-intelligence-machine-learning-development-dfeee0eabb5f

Regards,

Shafagat

Yes your right ! "it is just a beginning for a new world"

Following

Following

Advanced Machine Learning and Engineering

I see 4 questions in 1:

which artificial intelligence method do you think will be the leader in next years?

-its contextual, hence no objective answer

Is the AI going to be replaced by any other approach ?

-Question is what do you understand by AI? as per question you think AI is an approach which i would say AI is much more than approach.

What kind of improvements do you think will

happen in this field?

Intellugent autonomous systems all around. And way forward simulating EQ and TOB will be biggest breakthrogh

Do you think the jobs will take growing in AI as nowadays?

Why not, for the people who wants to upgrade themselves yes.

Reinforcement learning . still many works to be done and still many applications of this field is not done yet.

I think that another AI would replace existin one, namely, Active Inference by Karl Friston, already showed to be more powerful than reinforcement learning.

At scientific conferences, forums devoted to artificial intelligence, ethical standards, trust, automatisation of unfair and biased practices, poorly predictable human behaviour are often called the main problems. These factors will in the near future determine the speed and success of the implementation of new technologies.

Ethical norms, behavioural stereotypes differ for different countries, groups of people or even states and districts.

Laws and regulations of almost all countries contain items of positive and negative discrimination.

Public opinion may be sensitive or insensitive to manifestations of political correctness.

That looks like an abstract philosophy, but tomorrow may be a mandatory regulation for all systems with AI.

I think that the future artificial intelligence methods will be concentrated in development build-in trust and ethic internal control and verification.

up to my knowledge on AI right now, I see that the current AI methods, requiring too much computational efforts. although they are very effective in modeling complex systems, but in my opinion, AI will continue to proceed in next years, but with upgraded methods!

e.g, biological computing systems are supposed to be the future of computers and if the AI methods match themselves to the bio computers, they will make a big step forward in the world of science and engineering.

Hybrid Intelligence - "Man-computer symbiosis is an expected development in cooperative interaction between men and electronic computers." (http://groups.csail.mit.edu/medg/people/psz/Licklider.html), "Intellectual power, like physical power, can be amplified" (http://pespmc1.vub.ac.be/books/IntroCyb.pdf).

Agree with Milena Cukic - we are moving forward post-connectionism (after a long time)

https://www.researchgate.net/publication/285613401_The_Concept_of_Dominance_by_AA_Ukhtomsky_and_Anticipation

https://www.springer.com/cda/content/document/cda_downloaddocument/9783319194455-c2.pdf?SGWID=0-0-45-1515874-p177396342

Thank you Drago Indjic, Interesting! Another link is with Norbert Wiener elaboration of psychiatric illness as functional disorders of the memory (1948), so we are recycling the knowledge which was already there for so many years...

BCI is just another application, but we are still far away of everyday use. A lot of work must be done...

I hope, explainable AI methods

Naturel IA = Advence in neuroscience will inspert rechears to imrove ther method.

The future in the artificial intelligence will slightly a lot of problems during the next decade, but the important question is what is the amount of what will affect artificial intelligence on life and this will affect positively on our lives .. I think it's the future of artificial intelligence in a constant evolution, but every evolutionary there are positive and negative effects

Two trends I see in AI, Deep Learning are;

1. Uncertainty in Deep Learning - This is a really important topic of discussion as it is crucial to have an understanding about the uncertainty of a prediction/classification model in AI based applications. Assistant Prof. Yarin Gal from Oxford has done a lot of research in this area recently (while he was a PhD student at Cambridge) - http://mlg.eng.cam.ac.uk/yarin/thesis/thesis.pdf

2. Diversity aware Deep Learning/AI - I have seen a lot of articles in MIT Technology review, Tech Crunch related to this area. Recent incidents such as self driving car AIs having a bias towards white people over black people have sparked the discussion about this Diversity in AI. I have seen a recent EU project called WeNet (https://twitter.com/WeNetProject) which tries to address problems related to Diversity in AI. Prof. Daniel Gatica-Perez from EPFL & Idiap, Switzerland is leading the work on this front according to the web page.

random forest

deep learning , reinforcement learning , auto ML

it depends, but I think it will be neuroevolution methods and algorithms

It depends on kind of problem and the novel mathematical models. I read papers a lot and I've detected two tendences:

1) Deep Reinforcement Learning for development of videogames, new systems of learning in education, and image processing.

2) Deep Clustering and Collaborative Clustering for Big Data. I think that granular computing could be other option for Big Data problems.

Nevertheless, I've analyzed a new hope in AI: Neuromorphic Computation.

Following

I think we have come back to the starting point: the same reflections and questions that ask our grandfather Alan Turing. Separating the data and processing was right but not enough rebust unfortunately this restriction (imposed by the machine) channeled us. But in any problem we can not separate the data from their treatments. . .

1. Big Data

2. Data Mining

3. Deep Learning and particularly Convolutional Neural Network

I think that everything will be based in Phyton https://www.python.org/, tensor flow, https://www.tensorflow.org/, processing https://processing.org/, raspberry PI https://www.raspberrypi.org/, i also think that it would be interesting to mix AI with VR https://www.oculus.com/, AR and new media technologies to create awesome technological and artistic worlds that will open the doors of the future of mankind. some useful technologies and languajes for that, would be Pure Data https://puredata.info/, max msp https://cycling74.com/, apple developer tools (Xcode, Quartz Composer, and all developer tools) https://developer.apple.com/, adobe tools, https://www.adobe.com/, Autodesk tools, https://www.autodesk.eu/, openframeworks tools, https://openframeworks.cc/, Unity https://unity.com, Unreal engine, https://www.unrealengine.com/en-US/what-is-unreal-engine-4, Microsoft developer tools, https://www.microsoft.com/en-us/download/developer-tools.aspx, blender https://www.blender.org/, Emotiv https://www.emotiv.com/, Arduino, https://www.arduino.cc/, 3d printing, https://www.3dhubs.com/guides/3d-printing/, OpenGl, https://www.opengl.org/, Android, https://www.android.com/, java https://www.java.com/es/download/, C++ https://isocpp.org/, http://cpp.sh/, https://www.headsem.com/mejores-ide-para-programar-en-c/, javascript https://www.javascript.com/, Mixed reality https://www.microsoft.com/en-us/hololens, https://www.leapmotion.com/, i hope this will be useful.

Deep reinforcement learning (interacting with the environment )

When we say artificial intelligence we talk about two aspects : metaheuristic algorithm and Machine learning algorithm, for machine learning i think that Deep ANN is the suited one for almost all the application, but it still depend on the problem that you try to solve.

The danger in stating which AI method will be the leader in next years is to think that every problem is a nail and we just have a hammer. This has been the problem with the current emphasis on deep learning.

Each problem is unique and must be solved with the appropriate approach, methodology and tools that are adequate for the problem (just take as example the no free lunch theorem in ML). AI is not ML and includes a multiplicity of algorithms and heuristics that in some instances cannot be compared side by side.

AI methods encompass a wide range of methods with different prerequisites and purposes.

• Deep neural networks, for example, can very well reproduce and extend, but are pretty bad at searching: their "pure" performance on classical combinatorial optimization problems (e.g., NP-hard problems such as the TSP) is quite a letdown since they are not fundamentally designed for such a task.

• Optimization-based methods (e.g. branch-and-bound algorithms, heuristics and metaheuristics as defined in the traditional OR literature), on the other hand, require a well-defined objective and an efficient procedure to calculate it. The lack of definition of a good objective in a variety of contexts restricts the scope of application of these methods.

⟾ For the future, it is increasingly important for a disciplined development of our fields to 1) know a variety of tools from different domains instead of single approach, and more importantly to 2) recognize the type of situations in which these tools are needed.

Moreover, combining the benefits of different approaches is a fertile research area: e.g., many ML approaches also hide some optimization models in their core, and many pure optimization methods can largely benefit from ML to guide the search...

auto encoder

For me, its sucess corresponds to the fact that several problems are more difficult to solve by using the clasical persspective

Data Protection Regulations, such as EU GDPR, are driving AI methods to be more explainable and comprehensive for the end user, such as Deep Learning algorithms. AI may suffer from explanation need in next years, and translate the algorithms to normal human being might be a new topic to be explored in this area.

If the world is sane in the future, then the paradigm which provides the most accurate reproducible answers should be the champion.

https://odajournal.com/2019/03/21/some-machine-learning-algorithms-find-relationships-between-variables-when-none-exist-cta-doesnt

Many researchers active today still subscribe to the historical approach, which requires a quiver-full of analytic arrows (methods, algorithms) to handle all of the different types of targets (research designs and hypotheses). The idea that one simple algorithm can be adapted to hit the bulls eye of any target (application) is alien--but this misconception is not true.

https://odajournal.com/2017/04/18/what-is-optimal-data-analysis

https://odajournal.com/2013/09/19/optimal-data-analysis-a-general-statistical-analysis-paradigm

But, you see some AI techniques are not fast enough and they need a very high computation efforts in order to run the. we nee AI techniques which are capable of simultaneous and online processing.

so which method do you think that are applicable with this point of view?

There are many methods that can handle various types of data and give accurate results. But if you want me to tell you which methods are capable of giving online and efficient results, I believe that the door is open to any development. And despite the existence of efficient methods at this time , but it did not reach the desired goals .. I aspire to the emergence of new methodrs in AI completely different

I am thinking about the composition of genetic algorithms, fuzzy logic and neural networks.

I think genetic algorithms are developing more.

I think deep neural networks are developing more.

A couple of facts supporting my understanding above (14 March) are:

NSF:

NSF's 10 Big Ideas - https://www.nsf.gov/news/special_reports/big_ideas/index.jsp

Idea #1 is about Hybrid Intelligence - https://www.nsf.gov/news/special_reports/big_ideas/human_tech.jsp

DARPA:

AI Next Campaign is also about human-machine symbiosis (Towards this end, DARPA research and development in human-machine symbiosis sets a goal to partner with machines) -

https://www.darpa.mil/work-with-us/ai-next-campaign

White House:

Strategy 2: Develop effective methods for human-AI collaboration. Increase understanding of how to create AI systems that effectively complement and augment human capabilities. -

https://www.whitehouse.gov/wp-content/uploads/2019/06/National-AI-Research-and-Development-Strategic-Plan-2019-Update-June-2019.pdf?fbclid=IwAR3QZaL_1w3Y4I7pEva8J8qIfjc_azzbSnyBjD63dZO5xqMAwYUA_dQ1IQ0

McKinsey:

- https://www.mckinsey.com/business-functions/marketing-and-sales/our-insights/the-future-of-personalization-and-how-to-get-ready-for-it?cid=eml-app

Personalization is one of the Hybrid Intelligence scenarios - http://iacio.org/wp-content/uploads/2019/04/IAC-Journal-of-CIO-and-Digital-Innovation-2018-9.pdf

I tkhink deep neural network combined with genetic algorithm.

Ant colony optimisation (ACO)