The Neuroscience Center
BRAIN RESEARCH
POTENTIAL THERAPIES
New Drugs
Trophic Factors
Engineered Antibodies
Small Molecules and RNAs
Cell and Gene Therapy
New Drugs
trial and error
potency with receptors is tested in a test tube
reaction is then observed and drug altered from there
Ex. interact solely with target
ex. less side effects
Growth Factors
control the development and survival of specific groups of neurons.
can be used for:
antibodies engineered to modify the interactions and toxicity of misfolded proteins are the cause of many neurodegen diseases;
small molecules that take advantage of specific biochemical pathways;
interfering RNAs (RNAi) that reduce toxic levels of individual proteins; and
stem cells that could replace dead or dying neurons.
Find NGF
clone its genes
modify TF-regulated functions
might not cure but could improve symptoms or slow progression
ex. NGF slows the destruction of neurons that use acetylcholine.
prevented cell death and stimulated the regeneration and sprouting of damaged neurons that die in Alzheimer’s disease
memory-impaired rats had it, remebered mazes as well as normal adult rats
NGF also holds promise for slowing the memory deficits associated with normal aging.
due to the discovery of other TFs, In the future, Alzh’s disease, Park’s disease, and ALS may be treated with TF or their genes.
growth factor therapy
neutralizing molecules that stop or inhibit growth can help repair damaged nerve fiber tracts in the spinal cord. Using antibodies that override the effect of Nogo-A, a protein that inhibits nerve regeneration, Swiss researchers succeeded in getting some nerves of damaged spinal cords to regrow in rats and monkeys.
both improvements in their ability to walk and use their forepaw digits after spinal cord damage
Now clinical trials: injured spinal cord injury patients are being treated with anti-Nogo-A antibodies.
engineered Antibodies
IS- can vaccinate brain against proteins out/inside cells
used in alzh's
cons: imflammatory reaction, where patients brain reacts to antibodies against the proteins
engineer antibodies or fragments of antibodies that can bind to and alter the disease characteristics of specific proteins.
delivered as either proteins or as genes
preliminary results for Huntington’s, Parkinson’s, and Alzheimer’s diseases, as well as neurodegenerative disorders such as variant Creutzfeldt-Jakob disease (vCJD), known as prion diseases. (similar to Mad cow)
Drosophila had huntingtons genes, too weak and uncoordinated to break out of pupa
treated w/ gene for an anti-HD antibody, all of them emerge as young adults.
Small Molecules and RNAs
small-molecule drugs, such as antibiotics and anti-tumor drugs, alter brain damage processes
animal models reduce the neuronal damage in ALS, Huntington’s disease, and Parkinson’s disease
Thousands of small-molecule drug candidates can be tested using high-throughput screening,
hundreds/thousands Cpds tested for desired cellular effect
neurodegen
these diseases have misfolded ptns who aggregate
lasers measure for aggregation
pack prions and SMs into containers
machine scans the containers and reports which drugs broke down the clump
Alzheimer’s and prion diseases have recently been described using these methods.
would be nice to decrease frequency of production of prions or their constituents
RNA drugs remove the RNAs that code for damaging ptns
Mouse models of Huntington’s disease and ALS appear to have responded positively, which are delivered via gene therapies.
Cell and Gene Therapy
neuronal stem cells — unspecialized cells that give rise to cells with specific functions — in the brain and spinal cord of embryonic and adult mice.
can continuously produce all 3 major cell types of the brain: neurons; astrocytes, the cells that nourish and protect neurons; and oligodendrocytes, the cells that surround axons and allow them to conduct their signals efficiently.
someday be useful for replacing brain cells lost to disease.
study how to convert adult cells to stem cells
pharmacologically directed to replace damaged neurons tailored to a specific patient and disease.
animal models of human diseases have shown that gene transfer vectors can be effective in correcting at least some aspects of neurological disease. At this time, adeno-associated virus (AAV) and lentivirus seem to be the safest and most efficient vectors. These vectors are being used in clinical trials in patients with Parkinson’s and in some rare genetic diseases. Herpes simplex virus and adenovirus vectors have been evaluated in early-stage human trials for treating brain tumors.
KINDS OF RESEARCH
Animal Research
Sample Research Methods
Imaging
Gene Diagnosis
Animal Research
prevalent thru neuroscience
role of Nts discovered in rats and mice
rabbits and cat: vision and other senses
zebrafish: transparent eggs
sea slugs: learning and memory
Chemical Connections in the Nervous System
Treatments for brain disorders such as Parkinson’s disease and ADHD target the synapse.
human body donors, monkeys, rats and mice
tract tracing show pathways and neurons containing Nts in them (roadmap)
learn how pathways affected by diseases
done for death of ACh in Alzh's.
Parkinson’s disease, has emerged through studies with rabbits and mice.
Nobel Laureate Arvid Carlsson, revealed that the neurotransmitter dopamine was being depleted. Using pigeons, scientists then discovered that this neurotransmitter was highly concentrated in the basal ganglia, the part of the brain involved in motor function
concluded: Dp neurons in BG die in PD, limiting production of Dp
lead to discovery of L-dopa
Rats study drug addiction
could nonhuman species become addicted to drugs?
take these drugs compulsively.
acivated the dopamine neurons of the VTA, which communicates with the nucleus accumbens
this pathway is also activated by natural rewards, such as food, water, and sex, but drugs of abuse can take over the reward system by mimicking or blocking the function of neurotransmitters
drugs affect ability to learn
Learning and Memory
Nobel Laureate Eric Kandel started mammals, too complex to enable him to study basic memory processes. Answer: sea slug
uncovered how short- and long-term memories are retained.
certain stimuli resulted in a more robust protective reflex, a form of learning for the sea slug. Furthermore, the strengthened reflex could remain in place for days and weeks as a short-term memory. Additional work showed that a stronger synapse was responsible for the retention of this information.
Long-term memories form in a different way. Stronger stimuli activate genes, resulting in an increase of some proteins and a decrease in others. These changes ultimately lead to the growth of new synapses. After demonstrating that both short- and long-term memory in sea slugs involve the synapse, Kandel was able to illustrate that similar mechanisms are at work in mice and other mammals.
Critical Periods
study critical periods in the development of vision.
monkeys and cats for treatment for amblyopia, vision of one eye is greatly reduced because the eyes do not work well together, has the best outcome when it is started early in life, before the age of eight.
visual experiences guide Striate development
afterwards, circuits cannot be easily modified
lead to a lot of aswers, including that amblyopia cannot be cured in adults
For this work, neuroscientists David Hubel and Torsten Wiesel won the Nobel Prize in 1981.
recent:
mice for factors change the brain to prevent rewiring after a certain age
modifying/removing them allows for changes in vision later in life
vision restored in adult amblyotic mice
Gene Diagnosis
The exact number of human genes is uncertain and the functions of many genes are still unknown, but the current estimate is that humans have approximately 20,000–25,000 pairs of genes contained in these 23 chromosomes.
More than 7,000 disorders, including many that affect the brain and neurodevelopment, are mutations ("misspelling").
Genetic linkage studies,
families and large groups of strangers diagnosed with specific conditions
find the chromosomal location of many genes.
Newer: chromosome microarrays
look at overall chromosome makeup of a person and find out if segments of chromosomes, perhaps involving multiple genes, are missing (called deletions) or present in more than the usual amount (called duplications).
recently helped identify many types of rearrangements of chromosome structure and specific genes that are associated with developmental disabilities and neurological disorders.
lead to:
clarify diagnoses and improve understanding of the cause of symptoms, allowing physicians to optimize methods of prevention and treatment
prenatal testing
carrier status evaluations
sometimes, genetic analysis can be helpful in evaluating the malignancy of specific tumors and reactions to certain medications and treatments.
Tracking down Genes
3 major gene discoveries via mapping
HTT, the gene that is altered in patients diagnosed with Huntington’s disease;
RB1, which causes inherited retinoblastoma, a rare, highly malignant, childhood eye tumor that can lead to blindness and death;
X-linked gene DMD, responsible for Duchenne muscular dystrophy, a progressive muscle disease.
mapping techniques have shown that one condition may actually be due to mutations in any one of a group of genes.
Walker-Warburg syndrome, severe problems involving the brain, eyes, and muscles, leading to death in infancy or early childhood. at least five genes are known to be associated with this disease, others yet to be discovered.
forms of intellectual disability, previously referred to as mental retardation, are also due to genes that are not working properly. Gene mapping enabled doctors to find the FMR1 gene, which is abnormal in people diagnosed with fragile X syndrome, the most common cause of inherited intellectual disability in males. FMR1 is located on the X chromosome and is important for neuronal communication.
searching for genetic components to schizophrenia, bipolar disorder, alcoholism, but no luck so far.
people missing a certain segment of chromosome 22, due to 22q deletion syndrome, have a higher chance of developing mental illness. However, not all people with this chromosome deletion develop mental illness, nor do all people with mental illness have such a genetic finding.
genes from parents can undergo changes in fetal development so that a child might have a genetic alteration that is not found in either of his parents.
child could have a genetic condition that may potentially be passed on to offspring but was not necessarily inherited from his parents.
ex. a gene called LIS1 that helps tell the brain how to grow. People with mutations in the LIS1 gene have smoother brains than normal and may have seizures. In addition, severe intellectual disability is common. Their parents dont have it. Low % for siblings.
genetic basis of autism
gene changes associated with autism, specifically, with conditions that can include autism or autism-like features as symptoms.
tuberous sclerosis complex, mutations in genes TSC1 and TSC2, as well as Rett syndrome, associated with the MECP2 gene.
Chromosome abnormalities via microarray
Deletions of a certain portion of chromosome 16 can lead to a variety of neurological symptoms, including autism in some individuals.
no one gene to autism diagnosis
once genes are implicated in a disease, now very early interventional studies for some neurological conditions, such as Angelman syndrome and tuberous sclerosis complex.
why different people who carry mutated versions of the same gene, even within a family, can have different types or degrees of symptoms, or sometimes no symptoms at all.
New Testing Tech
uncover functional sequence of 20,000 genes (exome) and remaining associated DNA that is thought to influence or regulate these genes (together with the exome, this is called the genome).
such studies have revealed numerous types of genetic variants, making for more variability in human genes than initially recognized.
led to the identification of the MLL2 gene responsible for Kabuki syndrome, which causes congenital intellectual disabilities along with certain abnormal facial features.
distinctive condition but still hard to find out its genetic abnormalities
BRAIN IMAGING
very few pathological lesions are confined to a precise functional area.
Structural brain imaging developed about 30 years ago.
Electrophysiological techniques for monitoring neuronal activity are based on changes in the membrane potential of activated neurons.
Brain scanning techniques work by monitoring changes in energy metabolism required by activate neurons
energy for Aps supplied by oxidation of glucose.
Glucose and oxygen are delivered to the brain by the cerebral circulation.
By virtue of the neurovascular link, there is a local increase in cerebral blood flow in active areas.
This occurs very quickly. Modern neuroimaging devices measure these changes in local cerebral blood flow and use them as an index of neural activity.
Positron emission Tomography (PET)
first functional technique
one of the most important techniques for measuring blood flow or energy consumption in the brain
detection of radioactivity emitted when positrons undergo radioactive decay in the brain.
injection of radioactive tracers that are attached to compounds of biological interest (drugs that bind to NT Rs), then carry the radioisotope to different brain areas.
The radioisotope shows up in the brain in proportion to how hard local neurons are working.
Rings of detectors around the subject’s head record the timing and position of gamma particles emitted by the nuclear isotope as it traverses the brain and decays.
Computers build threedimensional images of changes in blood flow based on the amount of radiation emitted in different brain regions.
The more brain activity, the more vivid the picture that is created.
understand how drugs affect the brain
what happens while people are working on different activities, such as learning and using language.
understanding certain brain disorders, such as stroke, depression, and Parkinson’s disease
measure changes in the release of some neurotransmitters
correlation of particular neurotransmitter and a behavior or cognitive process
PET has revealed marked changes in the depressed brain.
can be used to produce maps of changes in local cerebral blood flow (CBF). Such measurements have led to the localisation in the human brain of sensory, motor and cognitive brain functions.
several disadvantages of PET, the major one being that it requires the injection of radioactive tracers.
many people cannot have a PET scan, ex.children and women of child-bearing age, and the number of measures taken during a scan are limited.
single photon emission computed tomography (SPECT)
similar to PET, but its pictures are not as detailed.
SPECT is much less expensive than PET because the tracers it uses break down at a slower rate and do not require a nearby particle accelerator, typical of those used in nuclear physics, to produce them.
Magnetic Resonance Imaging (MRI)
high-quality, three-dimensional image of organs and structures inside the body without X-rays or other radiation, MRIs are noninvasive and unsurpassed in the anatomical detail they show. MRIs tell scientists when structural abnormalities first appear in the course of a disease, how they affect subsequent development, and precisely how their progression correlates with mental and emotional aspects of a disorder. In some instances, they can even reveal minute changes that occur over time.
15-minute MRI procedure, a patient lies inside a massive, hollow, cylindrical magnet and is exposed to a powerful, steady magnetic field. Different atoms in the brain resonate to different frequencies of magnetic fields.
A background magnetic field lines up all the atoms in the brain.
Then a second magnetic field, oriented differently from the background field, is turned on and off many times a second; at certain pulse rates, particular atoms resonate to and line up with this second field.
When the second field is turned off, the atoms that were lined up with it swing back to align with the background field.
As they swing back, they create a signal that can be picked up and converted into an image.
Tissue that contains a lot of water and fat produces a bright image; tissue that contains little or no water, such as bone, appears black
MRI often combined with DTI
gray matter and white matter
can be used to provide very fine-grained images of brain structure, and a recent development called diffusion tensor imaging (DTI) permits detailed images of the white matter tracts of fibres that connect brain regions.
different MRI procedure can also assess the path of fiber tracts in the brain regions.
referred to as diffusion tensor imaging, takes advantage of diffusion rates of water, which tend to be higher along fiber tracts, to produce highresolution images of how areas may connect in the brain
MRI images can be constructed in any plane, and they are particularly valuable in studying the brain and spinal cord. The images reveal the precise extent of tumors rapidly and vividly and provide early evidence of potential damage from stroke, allowing physicians to administer proper treatments early, when they can have an impact.
Magnetic Resonance Spectroscopy (MRS)
related to MRI, uses the same machinery but measures the concentration of specific chemicals — such as neurotransmitters — in different parts of the brain instead of blood flow.
MRS also holds great promise: By measuring the molecular and metabolic changes that occur in the brain, this technique has already provided new information about brain development and aging, Alzheimer’s disease, schizophrenia, autism, and stroke.
Because it is noninvasive, MRS is ideal for studying the natural course of a disease or its response to treatment.
Functional Magnetic Resonance Imaging (fMRI)
One of the most popular neuroimaging techniques today is fMRI.
study primary sensory responses to cognitive activities.
noninvasive
temporal and spatial resolution
This technique compares brain activity under resting and active conditions.
allows for more detailed maps of brain areas underlying human mental activities in health and disease.
combines the high-spatial resolution, noninvasive imaging of brain anatomy offered by standard MRI with a strategy for
based on the difference in magnetic properties of oxyhaemoglobin and deoxygenated haemoglobin in blood (hence the signal in fMRI is called the Blood-OxygenationLevel-Dependent signal – BOLD).
Recent developments mean that even very brief thoughts or brain events (as little as one or two seconds in duration) can be measured. This is known as event-related fMRI.
How it works
increased neuronal activity leads to movements of ions that activate energy-requiring ion pumps, there is an increase in energy metabolism and oxygen consumption.
leads to an increase in deoxygenated haemoglobin and a decrease of the magnetic signal.
However increased oxygen consumption is followed within seconds by an increase in local cerebral blood flow.
The increase in cerebral blood flow exceeds the increase in oxygen consumption; there is therefore a relative increase in oxyhaemoglobin and the size of the signal. The exact mechanism of the increased cerebral blood flow is still unclear, but neurotransmitter–related signalling is now thought to be responsible.
detecting increases in blood oxygen levels when brain activity brings fresh blood to a particular area of the brain — a correlate of neuronal activity.
Magnetoencephalography (MEG)
recently developed technique that reveals the source of weak magnetic fields emitted by neurons
array of cylindershaped sensors monitors the magnetic field pattern near the patient’s head to determine the position and strength of activity in various regions of the brain.
In contrast with other imaging techniques, MEG can characterize rapidly changing patterns of neural activity — down to millisecond resolution — and can provide a quantitative measure of the strength of this activity in individual subjects.
Moreover, by presenting stimuli at various rates, scientists can determine how long neural activation is sustained in the diverse brain areas that typically respond. One of the most exciting developments in imaging is the combined use of information from fMRI and MEG.
The former provides detailed information about the areas of brain activity while an individual is engaged in a particular task, whereas MEG tells researchers and physicians when certain areas become active. Together, this information leads to a much more precise understanding of how the brain works in health and disease
Optical Imaging and Other Techniques
shining weak lasers through the skull to visualize brain activity.
These techniques are inexpensive and relatively portable.
They are also silent and safe: Because only extremely weak lasers are used, these methods can be used to study everyone, even infants.
near infrared spectroscopy (NIRS)
technicians shine lasers through the skull at near infrared frequencies, which renders the skull transparent.
Blood with oxygen in it absorbs different frequencies of light from blood in which the oxygen has been consumed.
By observing how much light is reflected back from the brain at each frequency, researchers can track blood flow.
Diffuse optical tomography is then used to create maps of brain activity.
A similar technique, the event-related optical signal, records how light scatters in response to rapid cellular changes that arise when neurons fire, potentially assessing neural activity lasting milliseconds.
transcranial magnetic stimulation (TMS),
works by inducing electrical impulses in the brain.
This is accomplished by altering magnetic fields through the use of an electromagnetic coil that emits powerful magnetic pulses while held against the scalp.
Repetitive TMS is being used to investigate the role of specific brain regions during behavior, and it can be combined with other neuroimaging techniques.
example, when TMS is used with fMRI, a functional correlation between a region and a behavior can be established.
Applications
subtraction methods
colour processing is in area V4, while motion processing (of random dots moving about on a screen) activates V5
statistical parametric mapping (SPM)
SPM maps are often given colours, with a fiery yellow used for the ‘hottest’ areas of activity through to blue and black for ‘cooler’ areas. Brain imaging scientists speak of areas ‘lighting up’ when certain functions are carried out. If a person watches a constantly changing checkerboard pattern, substantial activation is observed in the primary visual cortex. The use of moving and coloured colour patterns and other clever stimuli designed to activate different areas of the visual system has given us a great deal of new information about the organisation of the human visual system.
indentify brain areas of reading – such as transforming visual words into a phonological code, the grouping of phonemes into whole words, the process of extracting the meaning of words, and so on. Learning tasks have also been studied, including work dissociating the brain areas involved in anticipating and perceiving pain.
unexpected failure tosee the medial temporal lobe lighting up routinely in long term memory tasks. However, newer testing paradigms – some including virtual reality - are now revealing its activity in memory processing along with other areas such as the prefrontal cortex and precuneous. Coupled with new neuropsycholgical and other imaging findings, this diversity of brain areas involved has led to a revision of our understanding of the memory systems of the brain. New mathematical techniques are also being developed to look at how the neural activity of different brain regions interacts and correlates during complex tasks - known as effective connectivity). This measure allows us to appreciate how brain areas work as a team and not merely as isolated functional hot spots. The hope is that these new techniques, with magnets of high field strength providing even more precise images, will tell us about the dynamics of networks of neurons talking to each other in the seamless control of perception, thought and action.
ARTIFICIAL BRAINS AND NETWORKS
All real brains consist of highly interconnected neuronal networks. Their neurons need energy and the networks need space. Our brain contains roughly 100 billion nerve cells, 3.2 million kilometers of ‘wires’, a million-billion connections, all packed into a volume of 1.5 litres, but weighing only 1.5 kg and consuming a mere 10 watts. If we tried to build such a brain using silicon chips, it would consume about 10 megawatts, i.e. enough electricity to power a town.
heat produced by such a silicon brain would cause it to melt!
discover how brains operate so efficiently and economically, and to use similar principles to build brain-like machines.
Building brain circuits in silicon
The energy cost of signaling - from one neuron to another - has probably been a major factor in the evolution of brains. About 50-80% of the total energy consumption of the brain is consumed in the conduction of action potentials along nerve fibres and in synaptic transmission.
rest of energy to manufacturing and maintenance
making connections between even modest numbers of silicon neurons is limited by the two-dimensional nature of chips and circuit boards. So unlike the brain, direct communication between silicon neurons is severely restricted.
exploiting the very high speed of conventional electronics, the impulses from many silicon neurons can be ‘multiplexed’ - a process of carrying many different messages along the same wire. In this way, silicon engineers can begin to emulate the connectivity of biological networks.
To reduce power but increase speed, neurally-inspired engineers have adopted the biological strategy of using analogue rather than digital coding. Carver Mead, one of the ‘gurus’ of silicon valley in California, coined the description ‘neuromorphic engineering’ to describe the translation of neurobiology into technology. Instead of coding digitally in 0’s and 1’s, analogue circuits code in continuous changes in voltages, as do neurons in their sub-threshold state
calculations are then done in less steps
analogue computation can add, subtract, exponentials, integrate.
Because spike coding is energetically costly, efficient coding maximizes the amount of information represented in a pattern of spikes by reducing what is called redundancy. Energy efficiency is also increased by using as small a number of active neurons as possible. This is called sparse coding and it provides another important design principle for engineers building artificial neural networks.
A silicon retina
silicon retina that captures light and adapts its output automatically to changes in overall lighting conditions. It connects to two silicon neurons that, like real neurons in the visual cortex, have the job of extracting information about the angles of lines and contrast boundaries in the retinal image.
The neurons in this prototype are called integrate-and-fire neurons and neuromorphic engineers use them a lot.
They get this name because they ‘add up’ the weighted inputs, coded as voltages that are arriving at their synapses, and only ‘fire’ an action potential if the voltage reaches a set threshold.
The silicon neurons themselves are built of transistors, but instead of using the transistors as switches and driving the voltages to saturation as in conventional digital systems, the transistors are operated in their subthreshold range.
In this range, they act more like the cell membranes of real neurons.
Additional transistors provide active conductances to emulate the voltage- and time-dependent current flows of real ion channels. This small visual system is a prototype for much more elaborate artificial visual systems that are under development, but even it illustrates how a very noisy real-world input can be processed rapidly to produce a simple decision.
It can do what it is designed to do - tell the orientation of a line in a scene - and neuroscientists are already using this simple silicon visual system to test equipment and train students.
The most important things about artificial networks is that they operate in the real world, in real time and use very little power.
Artificial Neural Networks
Artificial neural networks (ANNs) are often used to study learning and memory. Usually they software on a conventional digital computer, they consist of a number of simple processing units that are highly interconnected in a network.
The simplest form of ANN is a feedforward associator, which has layers of interconnected input and output units.
An associative memory is encoded by modifying the strengths of the connections between the layers so that, when an input pattern is presented, the stored pattern associated with that pattern is retrieved.
A more complex ANN is a recurrent neural net.
This consists of a single layer where every unit is interconnected and all the units act as input and output. It sounds a bit strange, but this design enables the net to store patterns rather than merely pairs of items.
Decoding this kind of autoassociative network is achieved by a recursive search for a stored pattern. It has been shown that for a network of 1000 units, about 150 patterns can be retrieved before errors in the retrieval patterns become too large.
The similarity of ANNs to brains lies in the way they store and process information.
The ‘knowledge’ that they process resides in the network itself.
They have no separate memory location like the digital computer, for which the arithmetic processor and memory addresses are separate.
Instead, they have content-addressable storage.
In an ANN, information is stored in the weights of the connections, the same way that synapses change their strength during learning.
Nor are ANNs programmed to perform any given procedure. Each ‘neuron’ inside is ‘dumb’ and simply responds according to the sum of its weighted inputs.
Still, they can be trained to clever things. The learning rules that train networks do so by modifying the strength of the connections between the neurons, a common one being a rule that takes the output of the network to a given input pattern and compares it with the desired pattern. Any ‘error’ in the comparison is then used to adjust the weights of the connections to achieve a closer output to the desired one. The network gradually reduces the error signal to a minimum.
This works - but only slowly. Mistakes turn out to be important - no learning is possible if the network cannot make mistakes. This is a feature of learning that can get overlooked.
Over-trained networks that made no errors would end up responding only to one type of input.
Such networks are metaphorically called grandmothered - a reference to mythical ‘grandmother cells’ in the human brain that might respond only when one’s grandmother comes into view and must never make a mistake!
This is not very useful in real world applications because everything we had to learn would require a separate network.
On the contrary, the neat thing about ANNs lies in their ability to generalize to input patterns they have never been exposed to in training. They see relationships, capture associations and discover regularities in patterns.
And they are fault - tolerant just like real brains.
They can still retrieve a stored pattern even when the input pattern is noisy or incomplete.
These are very important properties for biological brains and ANNs can do these things too.
The paradox of modern computing technology
The paradox of present-day ANNs is that they are simulated mathematically on digital computers. This makes their use in real - world situations much more limited, because the simulation takes time and so the ANNs cannot operate in real time. ANNs might seem well-suited to drive an automobile, or fly an aircraft, because they are robust in the face of noise and keep going when some units in the network cease to work. However, the expert systems that are generally used in automatic pilots are digital computers programmed with conventional deterministic software and, for safety, this always require a backup. If things ever go badly wrong with the aircraft, such expert systems cannot cope. The human pilot must take over. Present-day training algorithms for ANNs are too slow for such emergencies. If silicon neurons could learn, which so far they can’t, then many of these problems would fall away. As we learn more about the way in which brains work, we will be able to build more sophisticated neural networks that will provide real brain-like performance.
Your brain is 100,000,000,000 cells and 3,200,000 kilometres of wires, with 1,000,000,000,000,000 synaptic connections, all packed into 1.5 litres and weighing 1.5 kg. Yet it consumes only about the same amount of electric power as a night-light!