We review 3^rd Waves of Artificial Intelligence (AI). The 1st Wave is 5 decades ago MIT Marvin Minsky with Seymour Papert proposed rule based “Perception”: "If so so, Then so so" system against the brain base “Perception” by Cornell Frank Rosenblat (1928-1971). The 2nd Wave is circa March 15 2016 "learn-able Rule Based system" with supervised learning with labeled data "from A to B" that Alpha-Go Brain had beat human (Korean genius Lee Sedol) in Go Chess Games (democratic black-white territorial game) 4 to 1. The revolution is due to (1) Massive Parallel Distributed (MPD) Compute (e.g. miniaturized GPU, as PC backplane), (2) matching MPD software (without the need of slow-down inner do-loop, Python: Tensor Flow, e.g. Stanford Course RA, MIT AGI), and (3) immense amount training data (e.g. in the Cloud, or Tesla Model 3-new Electrek collecting real world driving data from owners paid inexpensively at $35K). The paper describes 3rd Wave of AI that has human Fuzzy Linguistic Thinking using Unsupervised Learning of Homo sapiens at a constant temperature in the sensory pairs called the power of pairs, e.g. “Agree, the Signal; Disagree, Noise” in terms of two LIDAR’s, RADAR’s, Videos’ pairs operated by their relaxation toward the Minimum Helmholtz Free Energy.

In the 3^rd Wave AI, we are not making robots thinking like a human, but robots can recognize human imprecise (to computer) thinking. We are not demand the computer robot to have the human-like intuitive, artistic and creative thinking utilizing both logical side and emotional side of brain. For example, how can machine define in the linguistic Sense, e.g. "Beauty, or Beautiful’ or."Young, or Youthful", these emotional words which are all open sets of possibilities. They cannot be normalized as the unit probability, but the open set possibilities which could be infinite and are powerful human thinking. It may be referred to Fuzzy Membership Function Logic, (FMFL) as first introduced by the former Prof. Lotfi Zadeh (LZ passed away at 98 years old 2017), confided once to the author (HS) as a founder of International Neural Net Society (INNS) that the logic is not fuzzy but the membership is; but he dropped the MF while he was Chairman of ECE of UC Berkeley becoming FL for funding reasons) (Figure 1). However, it was late Prof. Walter Freeman (WF) at UC Berkeley (passed away at 89 years 2017) recognized the open possibility could not be loaded in the von Neumann finite state machine, namely computer, but Homo sapiens, namely our species, adopted both sides of brain in estimating approach, not by a look up table, but generating them by experience, (if you will unbounded table smart approach). Of course, we shall replace “human intuitive estimation” with the next best fast MPD computational ensemble approach averaged with all possible initial and boundary conditions. However, when the Boolean Logic "union & intersect" the Young with the Beautiful, the result becomes much sharper in the meaning that all human understood. An anecdote story, during one of the 30^th INNS board meeting, was between WF and HS the typical French Mathematician Peano’s theorem of real positive integer, that are not unbounded, non-denumerable, to be put in the finite state computer; but computational dynamical approach can, If it defines Real Positive Integer to be $I$ ,

Ifetc. Then Compute. Q.E.D.                                                 (1)

This task computer can easily handle.

According to the Science Magazine on December 15, 2017 weekly issue "When will we get there?" we found the Level 4 Automation of Autonomous Vehicle (AV) will take another 13 years.^1,2 (The final stage will be Level 5, with no human at all). In spite of early efforts of NASA space landing of Martian Cruiser, and DARPA Grant Challenge of Road Follower (Figure 2) that were able to differentiate a pot hole from tree shadows in a so-called “no-man land.”

Figure 1 Late Professors Lotfi Zadeh and Walter Freeman were associated with UC Berkeley and the lead Author HS at International Neural Network Society during the last 30 years.

Figure 2 NASA Martian Cruiser, and DARPA Road Follower , that can eventually differentiate a shadow from real pot-hole to go off the road; but aal have done in a no-man land.

The 1^st Wave AI was competing between MIT Marvin Minsky show that single layer cannot solve the EXOR logic and Cornell Frank Rosenblatt one-layer “perception” (Figure 3) (Figure 4). Modern Google and Volvo/Uber spent $ billions investment (since Volvo/Uber AV killed an intoxicated woman crossing intersection with pink pike in Temple Arizona), the Highway TSA committee nevertheless said it will take another 13 years for Automation Level 4 for Autonomous Vehicle (AV).^3,4 This might be due to that in current computer scientists thinking and designs of an AV are missing about “Human is creative animal we may abide the law but violate the rules. We suggest that one of the shortfalls might be that the computer scientist’s community is not familiar with "Orifice-like focusing logic" that begins with all possibility inputs and end up much focused result as follows.

Figure 3 The historical lesson initiated the multiple layer studies by MIT PDP book in backward error propagation called “Deep Learning, simulating Human Visual System V1(edge)-V2(shape)-V3(contour)-,V4 (Aided Target Recognition) layer by layer that have defeated Korean Genius Go Chess player.

Figure 4 The 2^nd Wave of AI can observed human play habit in a Learnable Rule Based System, e.g. Google Alpha Go that can anticipate rapidly many steps deep than human.

We shall compute all 5 dynamic equations with lab model car simulation results, e.g.

1. Newtonian dynamics of inertial centrifugal force in freeway tuning
2. Road-tire friction Langevin equation with variable road and weather conditions
3. Lyaponov Control theory equation at aforementioned MFE.,
4. Global Positioning Satellite Orbital Intersection Systems to increase to one feet resolution
5. Power of Pairs Sensors (Radar’s, LIDAR’s,Video’s) forming the situation awareness.

The guiding principle will be minimizing the Herman Helmholtz free energy H defined by the Boltzmann irreversible thermodynamics of the Entropy S (as the measure of the degree of uniformity, e.g. beach white sands have more entropy than mountain top rocks).Then, A.M. Lyaponov defined the monotonic decay control theory We ask how to make machine understand human. We assume the bases of homo sapiens natural survival intelligence (or in short Natural Intelligence (NI) as opposed to AI). NI assumes the homeostasis: namely two attributes are

1. Quick response with Power of Pairs sensors (two eyes, two ears, two nostrils, two sides of tongue) accomplish "While the agreed, the signal; disagreed, the noise"
2. Effortless unsupervised learning at Helmholtz's Minimum Free Energy (MFE) not supervised Least Mean Squares (LMS) errors. Anatomically speaking, brain consists of 10 billion neurons using Calcium ions as the communication vesicles positive current modulated at the synaptic gap junction with the help 10 times smaller and 10 times more (about 100 billion) of house-keeping glial cells, keeping the ions in the line of axon, “While one pops in, the other pops out instantaneously”. What is the internal energy of brain that is free to do the work? According to Ludwig Boltzmann, the measure of useless thermal energy is called the entropy when multiplied with the Kelvin temperature".

$TotalEntropy=S_{brain}+S_{reservoir}=K_{B}LogW$;   (1)

$W=\exp \left(\frac{S_{brain+S_{reservoir}}}{}{K}_B\right)=\exp \left[\frac{S_{brain+S_{reservoir}}}{K_B{T}_0}\right]=exp\left[S_{brain}{T}_o-E_{brain}\right)/k_B{T}_o=exp\left[-H_{brain}/k_B{T}_o\right]$   (2)

Herman Helmholtz has the first derived from canonical ensemble from the thermodynamic free energy; we applied it to the human brain

$H_{brain} \equiv E_{brain} –S_{brain}{T}_o$    (3)

Use is made of the conservation of energy the brain internal energy$\Delta E_{brain} +\Delta S_{reservoir}{T}_o=0$ in equilibrium with the thermal reservoir heat energy $\Delta S_{reservoir}{T}_o$ . We can drop the integration constant as the normalization constant. When the input pair sensors agree, the stimuli are added up as a double in the brain. It shall be relaxed to the base noise level at constant temperature.

Homeostasis: human $T_o = {37}^{o}C$ , chicken ${40}^{o}C$ ; but the hotter is not any smarter;

Prof: "we ate chicken, not vice versa, Q.E.D.")

Thus, we have arrived at the MFE as the guiding principle of human being.

A.M Lyaponov control theory follows

$min.H\downarrow =E\downarrow -T_{o}S\uparrow$    (4)

$\frac{\Delta H}{\Delta t}=\frac{\Delta H}{\Delta E}\frac{\text{Δ}E}{\text{Δ}t}=-{\left(\frac{\Delta H}{\Delta E}\right)}^2\le 0;$    (5)

$f and only if \frac{\text{Δ}E}{\text{Δ}t}=-\frac{\Delta H}{\Delta E}$ Newtonian dynamics    (6)

Similar to the botanist Brown observing the mysterious pollen ceaseless zigzag motions in the water pound, Albert Einstein understood that he saw by his naked eyes the molecules thermal motion and he emulate the smoke particles kicking randomly by air molecules in the diffusion trajectory in the air (Figure 5). Mean square displacement is no longer quadratic but linear

Figure 5 Brownian motion is similar to UAT in a road test of microscopic road-tire friction and weather condition vaying. We shall take the envelop trajectory.

$\langle \stackrel{\rightharpoonup }{r}\left(t\right)\cdot \stackrel{\rightharpoonup }{r}\left(t+\tau \right)\rangle ={{\displaystyle \iint }}^{\text{}}\langle \stackrel{\rightharpoonup }{v}\left(t\right)\cdot \stackrel{\rightharpoonup }{v}\left(t+\tau \right)\rangle dt d\left(t+\tau \right)=\frac{D}{k_B{T}_o\mu }{v}^2\delta \left(\tau \right)$    (7)

We compute all five dynamic equations on the fly, and the result is tabulated into 5 Fuzzy Membership Functions (FMF’s). Then the Boolean Logic at the decision time at a red traffic light in a desert town, at the midnight the AV shall slow down and glide over the red light. “A rule is made to break in a truly human intelligence behavior".

We believe that the multiple FMF's approach to simulation data might be the missing sciences that can shorten the sensible AV results from 13 years to time 6.5 years without the violation of the Cardinal rules; "Thou shall do no harm to human". Otherwise, the impatient human driver might take over the control of AV, and coast over slowly through a red light in the midnight of desert.

The impact area will be Unmanned Autonomous Truck (UAT) boosting the US economics.

There have been historically three economic impacts, namely'

1. The 1^st economic Booming: President Eisenhower built the Interstate Highways for missile launcher , but stimulated Interstate Goods exchanges
2. The 2^nd economic Booming: President John F. Kennedy to compete against Russian satellite landed human on the moon developing the precision electronics and feedback control theory those producing high power computers.
3. The 3^rd economic Booming, Vice-President Al Gore recognized ARPA internet linked all PI'S to PM, to all agencies, to all societies.
4. The 4th economic Booming: we anticipate being Unmanned Autonomous Trucks (UAT) Figure 6 that link highway depots to depots, while drivers become depot managers having more salary.

Figure 6 Unmanned Autonomous Trucks (UAT) from highway fixed depots to depots while drives working as depot managers to save the most costly element human drivers ( we believe Amazon might not need East Coast Head Quarter at National Landing Crystal City , and other Supermarket chain stores etc. can all reduce the cost of retail goods). Thus, an early development of a scaled down model conducting the road test in a room should be shortened in 3~5 years.

We acknowledged ONR Code 321 initial seed money to make this planning possible. Experimental simulation results will be finalized and published elsewhere. This is a short proposal to precede multiple universities and multiple year’s collaboration, under the guidance of DARPA Dir.Dr. Steven Walker unveiled $2B effort to develop the next wave of AI.^5,6

Authors declare that there is no conflict of interest.

1. Jeff Mervis. When will we get there? Science V. 2017. p. 1370–1375.
2. Yann LeCun, Yoshua Bengio, Geoffrey Hinton. Deep Learning. Nature V. 2015. p. 436–444.
3. https://www.motor1.com/news/237427/uber-volvo-autonomous-tech-crash/
4. Many Automobile companies including Tesla, Toyota, Volvo have spent at the tone of multiple billions dollars investment on Autonomous Vehicle; unfortunately one subsidiary of Volvo/Uber have accidently killed an (intoxicated?) woman pushing a pink pike crossing the traffic intersection in Temple Arizona. Insurance paid off the dead woman; but suited the car design company which has violated the fundamental.
5. Analternative to the Hippocratic Oath for Physician: “Thou shall do no harm to human is not give up on patients” for the future robot machine physician society.  This is one of the ethic rules that are of concern to the professional society, such as ACM, IEEE/CIS, etc.
6. DARPA director, Dr. Steven Walker, officially unveiled Announces $2 Billion Campaign to Develop Next Wave of AI... https://www.darpa.mil› News And Events Sep 7, 2018  Opportunities. DARPAsees this next generationofAI as a third waveof technological advance, one of contextual adaptation called the “AI Next” campaign.