New Hampshire House takes on artificial intelligence in political advertising

Updated: 3:44 PM EDT Mar 29, 2024

Political advertisements featuring deceptive synthetic media would be required to include disclosure language under a bill passed Thursday by the New Hampshire House.Sophisticated artificial intelligence tools, such as voice-cloning software and image generators, are already in use in elections in the U.S. and around the world, leading to concerns about the rapid spread of misinformation.In New Hampshire, authorities are investigating robocalls sent to thousands of voters just before the Jan. 21 presidential primary that featured an AI-generated voice mimicking President Joe Biden. Steve Kramer, a political consultant, later said he orchestrated the calls to publicize the potential dangers of artificial intelligence and spur action from lawmakers. But the attorney general's office has said the calls violated the state's voter suppression law.The bill sent to the Senate on Thursday would require disclosure when deceptive artificial intelligence is used in political advertising within 90 days of an election. Such disclosures would explain that the advertisings image, video or audio has been manipulated or generated by artificial intelligence technology and depicts speech or conduct that did not occur.The bill, which passed without debate, includes exemptions for satire or parody.

Political advertisements featuring deceptive synthetic media would be required to include disclosure language under a bill passed Thursday by the New Hampshire House.

Sophisticated artificial intelligence tools, such as voice-cloning software and image generators, are already in use in elections in the U.S. and around the world, leading to concerns about the rapid spread of misinformation.

In New Hampshire, authorities are investigating robocalls sent to thousands of voters just before the Jan. 21 presidential primary that featured an AI-generated voice mimicking President Joe Biden. Steve Kramer, a political consultant, later said he orchestrated the calls to publicize the potential dangers of artificial intelligence and spur action from lawmakers. But the attorney general's office has said the calls violated the state's voter suppression law.

The bill sent to the Senate on Thursday would require disclosure when deceptive artificial intelligence is used in political advertising within 90 days of an election. Such disclosures would explain that the advertisings image, video or audio has been manipulated or generated by artificial intelligence technology and depicts speech or conduct that did not occur.

The bill, which passed without debate, includes exemptions for satire or parody.

Continued here:
New Hampshire House takes on artificial intelligence in political advertising - WMUR Manchester

Hank Stephenson has a finely tuned B.S. detector. The longtime journalist has made a living sussing out lies and political spin.

But even he was fooled at first when he watched the video of one of his home states most prominent congressional candidates.

There was Kari Lake, the Republican Senate hopeful from Arizona, on his phone screen, speaking words written by a software engineer. Stephenson was watching a deepfake an artificial-intelligence-generated video produced by his news organization, Arizona Agenda, to underscore the dangers of AI misinformation in a pivotal election year.

When we started doing this, I thought it was going to be so bad it wouldnt trick anyone, but I was blown away, Stephenson, who co-founded the site in 2021, said in an interview. And we are unsophisticated. If we can do this, then anyone with a real budget can do a good enough job that itll trick you, itll trick me, and that is scary.

As a tight 2024 presidential election draws ever nearer, experts and officials are increasingly sounding the alarm about the potentially devastating power of AI deepfakes, which they fear could further corrode the countrys sense of truth and destabilize the electorate.

There are signs that AI and the fear surrounding it is already having an impact on the race. Late last year, former president Donald Trump falsely accused the producers of an ad, which showed his well-documented public gaffes, of trafficking in AI-generated content. Meanwhile, actual fake images of Trump and other political figures, designed both to boost and to bruise, have gone viral again and again, sowing chaos at a crucial point in the election cycle.

Now some officials are rushing to respond. In recent months, the New Hampshire Justice Department announced it was investigating a spoofed robocall featuring an AI-generated voice of President Biden; Washington state warned its voters to be on the lookout for deepfakes; and lawmakers from Oregon to Florida passed bills restricting the use of such technology in campaign communications.

And in Arizona, a key swing state in the 2024 contest, the top elections official used deepfakes of himself in a training exercise to prepare staff for the onslaught of falsehoods to come. The exercise inspired Stephenson and his colleagues at the Arizona Agenda, whose daily newsletter seeks to explain complex political stories to an audience of some 10,000 subscribers.

They brainstormed ideas for about a week and enlisted the help of a tech-savvy friend. On Friday, Stephenson published the piece, which included three deepfake clips of Lake.

It begins with a ploy, telling readers that Lake a hard-right candidate whom the Arizona Agenda has pilloried in the past had decided to record a testimonial about how much she enjoys the outlet. But the video quickly pivots to the giveaway punchline.

Subscribe to the Arizona Agenda for hard-hitting real news, the fake Lake says to the camera, before adding: And a preview of the terrifying artificial intelligence coming your way in the next election, like this video, which is an AI deepfake the Arizona Agenda made to show you just how good this technology is getting.

By Saturday, the videos had generated tens of thousands of views and one very unhappy response from the real Lake, whose campaign attorneys sent the Arizona Agenda a cease-and-desist letter. The letter demanded the immediate removal of the aforementioned deep fake videos from all platforms where they have been shared or disseminated. If the outlet refuses to comply, the letter said, Lakes campaign would pursue all available legal remedies.

A spokesperson for the campaign declined to comment when contacted on Saturday.

Stephenson said he was consulting with lawyers about how to respond, but as of Saturday afternoon, he was not planning to remove the videos. The deepfakes, he said, are good learning devices, and he wants to arm readers with the tools to detect such forgeries before theyre bombarded with them as the election season heats up.

Fighting this new wave of technological disinformation this election cycle is on all of us, Stephenson wrote in the article accompanying the clips. Your best defense is knowing whats out there and using your critical thinking.

Hany Farid, a professor at the University of California at Berkeley who studies digital propaganda and misinformation, said the Arizona Agenda videos were useful public service announcements that appeared carefully crafted to limit unintended consequences. Even so, he said, outlets should be wary of how they frame their deepfake reportage.

Im supportive of the PSAs, but theres a balance, Farid said. You dont want your readers and viewers to look at everything that doesnt conform to their worldview as fake.

Deepfakes present two distinct threat vectors, Farid said. First, bad actors can generate false videos of people saying things they never actually said; and second, people can more credibly dismiss any real embarrassing or incriminating footage as fake.

This dynamic, Farid said, has been especially apparent during Russias invasion of Ukraine, a conflict rife with misinformation. Early in the war, Ukraine promoted a deepfake showing Paris under attack, urging world leaders to react to the Kremlins aggression with as much urgency as they might show if the Eiffel Tower had been targeted.

It was a potent message, Farid said, but it opened the door for Russias baseless claims that subsequent videos from Ukraine, which showed evidence of Kremlin war crimes, were similarly feigned.

I am worried that everything is becoming suspect, he said.

Stephenson, whose backyard is a political battleground that lately has become a crucible of conspiracy theories and false claims, has a similar fear.

For many years now weve been battling over whats real, he said. Objective facts can be written off as fake news, and now objective videos will be written off as deep fakes, and deep fakes will be treated as reality.

Researchers like Farid are feverishly working on software that would allow journalists and others to more easily detect deepfakes. Farid said the suite of tools he currently uses easily classified the Arizona Agenda video as bogus, a hopeful sign for the coming flood of fakes. However, deepfake technology is improving at a rapid rate, and future phonies could be much harder to spot.

And even Stephensons admittedly sub-par deepfake managed to dupe a few people: After blasting out Fridays newsletter with the headline Kari Lake does us a solid, a handful of paying readers unsubscribed. Most likely, Stephenson suspects, they thought Lakes endorsement was real.

Maegan Vazquez contributed to this report.

View post:
AI generated deepfake of Kari Lake used to highlight dangers in election - The Washington Post

A micropore device

Micropores (3m in diameter) were fabricated on a 50nm-thick silicon nitride film cast on a silicon substrate. The silicon substrate was sandwiched between a 25mm25mm 0.5-mm-thick plastic channel. This structure is termed a micropore module or a module (Fig.1a, b). Bacterial suspensions (18 L) in 1phosphate buffered saline (PBS) and PBS (15 L) were introduced into the cis and trans chambers, respectively. We measured ionic currenttime waveforms through applying a voltage of 0.1V between the Ag/AgCl electrodes placed in the flow channel. When the bacteria passed through the micropores, the ionic current decreased owing to obstruction due to the flowing ions (Fig.1c). S. aureus and S. epidermidis are spherical in shape with a diameter ranging from 0.6 to 1.0m under a low vacuum of 98hPa. We observed no specific differences in the bacterial structures through scanning electron microscopy (SEM) under the culture conditions (Fig.1dg). The waveforms were collected in the data server and analysed, as described in the Methods section. The number of available waveforms for the bacterial suspensions was independent of their optical densities measured at a wavelength of 600nm (OD600) (Fig.1h). Therefore, sufficient bacteria were present in the suspensions regardless of their OD600 values.

(a) The structure of the micropore module. The micropore module is 25 mm2 in size and 0.5mm thick. The bacterial suspension is introduced from the cis channel, and PBS is introduced from the trans channel. (b) An optical image of the pore of the module. The optical microscopic examination of the micropore module suggests that the 3m diameter micropore is in the centre of the silicon substrate. (c) A schema of the micropore. When the bacteria pass through the micropore, the ionic current decreases because of obstruction in flowing ions. The processing software obtains the change in ionic current as a waveform. (d) Scanning electron microscope (SEM) observation of S. epidermidis14,000. (e) SEM observation of S. epidermidis60,000. (f) SEM observation of S. aureus14,000. (g) SEM observation of S. aureus60,000. (h) A scatter diagram of the OD600 of the bacterial suspensions and their average pulse counts. No correlations were observed between the OD600 and pulse counts of the bacterial suspensions.

Ionic current measurements were performed on S. aureus and S. epidermidis cultures using micropore devices placed under an optical microscope. The base current was approximately 0.4 A. Using the equations for access resistance (1/D) and micropore resistance (4L/D2), the base current was estimated to be 0.5 nA when the ionic conductivity () of 1PBS was 1.61 Sm1, the average diameter of the micropores was D=3m, and the thickness of the micropores was L=50nm. The ionic current obtained was consistent with the theoretically estimated ionic current. With application of a voltage of 0.1V, a single bacterium (black spot indicated by yellow arrow, Fig.2a) was pulled into the micropore. Bacteria within an approximate radius of 15m from the micropore were pulled in at an accelerated rate as they approached the micropore (Fig.2a(langle 1rangle)). When one bacterium passed through the micropore, we observed a single ionic currenttime waveform, and the ionic current did not change until it entered the micropore. Bacteria within an approximate radius of 40m from the micropore were drawn to the micropore by Brownian motion (Fig.2a(langle 2rangle)). The duration of Brownian motion was>1min while approaching the micropores. When a bacterium reached an approximate 15m radius from the micropore threshold, it was rapidly pulled into the micropore, which suggested it was being pulled under the influence of electric forces.

(a) Optical microscope images (scale bar=10m) of S. epidermidis (small dot indicated by yellow arrow) being drawn towards and pulled into a micropore (3m black circles):<1>the bacterium near (within 15m radius) the micropore is pulled in at a fast rate;<2>the trajectory of a distant bacterium being drawn towards the micropore until it is pulled in. (b) Actual display screen during the measurement. The moment at which the bacteria (yellow arrow) is being sucked into the micropore. Its waveform is denoted on the right top column of the right-side window. The waveforms are indicated at three different time scales using the waveform viewer software.

Negatively charged bacteria were subjected to diffusion and electric forces between the electrodes. The radius (r) from the centre of the micropore at which the bacteria were trapped by the electric field is denoted by the equation:

$$r = d^{{2}} mu Delta V/{8}hD$$

where d, , and V are the diameter of the micropore, the mobility of the bacteria, and applied voltage, respectively, and h and D are the thickness of the micropores and diffusion constant of the bacteria, respectively. The bacterial mobility and diffusion constants were approximately108mV1s1 and 109 m2 s1, respectively. The experimentally observed r=15m was relatively close to the theoretically predicted r=22.5m. When the bacteria did not pass through the micropores, the ionic current remained constant at the base current, which was correlated with a non-event (Supplementary Material: the captured video of the actual screen during the measurement of the ionic currents of the bacteria. The optical microscopic image is shown on the left side, and the waveforms of ionic currents are shown on the right side). When a bacterium passed through the micropore, we obtained a spike-shaped ionic currenttime waveform corresponding to one bacterium (Fig.2b). The combination of optical microscopy and ionic current measurements demonstrated a correlation between the movement of a single bacterium and the ionic currenttime waveform.

Spike-shaped waveforms are characterised by a maximum current value (Ip) and current duration (td) (Fig.1c). The histograms of Ip and td nearly overlapped completely (Fig.3a, b). Similar Ip levels reflected a small difference in the size between the two bacterial species. It was difficult to distinguish between them using these histograms. However, we observed differences in the shapes of the waveforms between the bacteria, suggesting distinguishable features (Fig.3c). We used machine learning for identifying the waveform features such that the waveforms obtained could be used to identify the two species.

(a) Histograms of Ip and td of S. epidermidis. (b) Histograms of Ip and td of S. aureus. (c) Differences in the absolute value of the histograms of Ip and td between S. epidermidis and S. aureus. (d) Confusion matrix of all isolates. (e) Confusion matrix of the assembly machine learning results.

Fifty isolates of S. aureus and S. epidermidis were used for ionic current measurements. Each bacterial isolate was measured in triplicate using a micropore device, and each measurement lasted for 3min. The 15,000 waveforms obtained from the measurements were provided as inputs for the machine-learning training set. The accuracy of differentiating between the two species in a single waveform was F-value=0.59, which exceeded the F-value=0.5 for random discrimination (Fig.3d). In this single-waveform learning, it is determined which species each single waveform belongs to. The F-value is denoted by the harmonic mean of the sensitivity and precision, 2/(1/sensitivity+1/precision), as follows:

$$Sensitivity= frac{True , Positive}{True , Positive+False , Negative}$$

$$Precision=frac{True , Positive}{True , Positive+False , Positive}$$

$$F-measure=frac{2times Sensitivity , times , Precision}{Sensitivity , + , Precision}$$

For S. epidermidis, the confusion matrix yielded a sensitivity and precision of 9781/(9781+5219)=0.65 and 9781/(9781+6968)=0.58, respectively. This resulted in an F-value of 0.62. Similarly, the sensitivity and precision for S. aureus were 0.61 and 0.54, respectively, yielding an F-value of 0.57. The overall adopted F-value was the average of the F-values obtained for the two species. We determined the species-level accuracy of the bacterial identification on an isolate-by-isolate basis using machine learning. Of the 50 bacterial isolates, 25 were selected randomly as the training set for each species, respectively. We determined the training set yielding the highest F-value. Using the selected training set, we performed assembly learning to develop a classifier to determine whether a bacterial isolate belonged to S. aureus or S. epidermidis. In this assembly learning, single waveforms of one isolate/strain are treated as an aggregated data. Therefore, this assembly learning is an isolate/strain-focused learning. In contrast to the single-waveform learning, it is determined which species the aggregated data of each isolate/strain belongs to. In addition, the assembly machine learning uses the entire distribution of waveforms in each species, in addition to independent parameters (Ip, td, current vector, and time vector). The F-value was 0.93 (Fig.3e).

The classifiers created during the machine learning process were used to distinguish the remaining 25 bacterial isolates (Fig.4a) and the additional ATCC standard strains. Each isolate was assessed by the classifier for three measurements, and two or three correct responses from the three trials were regarded as the final correct answers for the isolate or strain.

(a) Hold-out method. We employed a hold-out method for machine learning that splits data into the following two groups: a training dataset and a testing dataset. (b) Receiver operating characteristic curve of the classifier. (c) Characteristic distribution of the waveform. The pulse data are acquired 250,000 times per min. Steps denote the number of data points acquired, Height denotes the current value of the pulse, and Peak ratio denotes the location of the peak of the pulse, when the left edge of the pulse is 0 and the right edge of the pulse is 1. (d) Zeta potential distribution of the bacteria. The dashed line denotes the median, and the dotted line denotes the quartiles. No statistically significant difference was noted between the two species.

The area under the receiver operating characteristic curve (AUROC) was 0.94 (>0.9), demonstrating that the trained classifier could distinguish S. aureus from S. epidermidis at a high accuracy (Fig.4b)20. The sensitivity and specificity for S. aureus detection were 96.4% and 80.8%, respectively, with an accuracy of 88.9% (Table 1). The positive agreement was 84.4%, and the negative agreement was 95.5% (Table 1).

Ionic currenttime waveforms consist of information in relation to the size, shape, and surface charges of bacteria passing through the micropores. Micropore measurements demonstrated few statistically significant differences in the size and shape of S. aureus and S. epidermidis (Fig.4c).

The Zeta potential affects the ionic current of the particles measured using the micropore device, and indicates the electrical charge of the surface layer of the particles21. Zeta potentials of the S. aureus and S. epidermidis isolates were measured using a Zetasizer (Malvern Instruments, Worcestershire, UK)22. The surface charges of the two species indicated that they were negatively charged, with Zeta potentials>20mV. While the Zeta potential range of S. epidermidis was greater than that of S. aureus and the distribution pattern was not completely the same, there were no significant difference between the Zeta potentials of S. aureus and S. epidermidis (Fig.4d). Machine learning used surface charge differences between the two bacteria to distinguish between the species in reference to the entire distribution pattern of the features, of which difference is too subtle to statistically detect. The ion currenttime waveform provides information on the volume, structure, and surface charge of bacteria passing through micropores. The machine learning, which inputs the shape of the ion currenttime waveform as a feature, is considered to distinguish differences in surface charge.

See the original post:
Using novel micropore technology combined with artificial intelligence to differentiate Staphylococcus aureus and ... - Nature.com

By The Henryk Niewodniczanski Institute of Nuclear Physics, Polish Academy of Sciences March 24, 2024

AI is emerging as a key tool in nuclear physics, offering solutions for the data-intensive and complex task of particle track reconstruction. Credit: SciTechDaily.com

Particles colliding in accelerators produce numerous cascades of secondary particles. The electronics processing the signals avalanching in from the detectors then have a fraction of a second in which to assess whether an event is of sufficient interest to save it for later analysis. In the near future, this demanding task may be carried out using algorithms based on AI.

Electronics has never had an easy life in nuclear physics. There is so much data coming in from the LHC, the most powerful accelerator in the world, that recording it all has never been an option. The systems that process the wave of signals coming from the detectors therefore specialize in forgetting they reconstruct the tracks of secondary particles in a fraction of a second and assess whether the collision just observed can be ignored or whether it is worth saving for further analysis. However, the current methods of reconstructing particle tracks will soon no longer suffice.

Research presented in the journal Computer Science by scientists from the Institute of Nuclear Physics of the Polish Academy of Sciences (IFJ PAN) in Cracow, Poland, suggests that tools built using artificial intelligence could be an effective alternative to current methods for the rapid reconstruction of particle tracks. Their debut could occur in the next two to three years, probably in the MUonE experiment which supports the search for new physics.

The principle of reconstructing the tracks of secondary particles based on hits recorded during collisions inside the MUonE detector. Subsequent targets are marked in gold, and silicon detector layers are marked in blue. Credit: IFJ PAN

In modern high-energy physics experiments, particles diverging from the collision point pass through successive layers of the detector, depositing a little energy in each. In practice, this means that if the detector consists of ten layers and the secondary particle passes through all of them, its path has to be reconstructed on the basis of ten points. The task is only seemingly simple.

There is usually a magnetic field inside the detectors. Charged particles move in it along curved lines and this is also how the detector elements activated by them, which in our jargon we call hits, will be located with respect to each other, explains Prof. Marcin Kucharczyk, (IFJ PAN) and immediately adds: In reality, the so-called occupancy of the detector, i.e. the number of hits per detector element, may be very high, which causes many problems when trying to reconstruct the tracks of particles correctly. In particular, the reconstruction of tracks that are close to each other is quite a problem.

Experiments designed to find new physics will collide particles at higher energies than before, meaning that more secondary particles will be created in each collision. The luminosity of the beams will also have to be higher, which in turn will increase the number of collisions per unit time. Under such conditions, classical methods of reconstructing particle tracks can no longer cope. Artificial intelligence, which excels where certain universal patterns need to be recognized quickly, can come to the rescue.

The artificial intelligence we have designed is a deep-type neural network. It consists of an input layer made up of 20 neurons, four hidden layers of 1,000 neurons each and an output layer with eight neurons. All the neurons of each layer are connected to all the neurons of the neighboring layer. Altogether, the network has two million configuration parameters, the values of which are set during the learning process, describes Dr. Milosz Zdybal (IFJ PAN).

The deep neural network thus prepared was trained using 40,000 simulated particle collisions, supplemented with artificially generated noise. During the testing phase, only hit information was fed into the network. As these were derived from computer simulations, the original trajectories of the responsible particles were known exactly and could be compared with the reconstructions provided by the artificial intelligence. On this basis, the artificial intelligence learned to correctly reconstruct the particle tracks.

In our paper, we show that the deep neural network trained on a properly prepared database is able to reconstruct secondary particle tracks as accurately as classical algorithms. This is a result of great importance for the development of detection techniques. Whilst training a deep neural network is a lengthy and computationally demanding process, a trained network reacts instantly. Since it does this also with satisfactory precision, we can think optimistically about using it in the case of real collisions, stresses Prof. Kucharczyk.

The closest experiment in which the artificial intelligence from IFJ PAN would have a chance to prove itself is MUonE (MUon ON Electron elastic scattering). This examines an interesting discrepancy between the measured values of a certain physical quantity to do with muons (particles that are about 200 times more massive equivalents of the electron) and predictions of the Standard Model (that is, the model used to describe the world of elementary particles). Measurements carried out at the American accelerator centre Fermilab show that the so-called anomalous magnetic moment of muons differs from the predictions of the Standard Model with a certainty of up to 4.2 standard deviations (referred as sigma). Meanwhile, it is accepted in physics that a significance above 5 sigma, corresponding to a certainty of 99.99995%, is a value deemed acceptable to announce a discovery.

The significance of the discrepancy indicating new physics could be significantly increased if the precision of the Standard Models predictions could be improved. However, in order to better determine the anomalous magnetic moment of the muon with its help, it would be necessary to know a more precise value of the parameter known as the hadronic correction. Unfortunately, a mathematical calculation of this parameter is not possible. At this point, the role of the MUonE experiment becomes clear. In it, scientists intend to study the scattering of muons on electrons of atoms with low atomic number, such as carbon or beryllium. The results will allow a more precise determination of certain physical parameters that directly depend on the hadronic correction. If everything goes according to the physicists plans, the hadronic correction determined in this way will increase the confidence in measuring the discrepancy between the theoretical and measured value of the muons anomalous magnetic moment by up to 7 sigma and the existence of hitherto unknown physics may become a reality.

The MUonE experiment is to start at Europes CERN nuclear facility as early as next year, but the target phase has been planned for 2027, which is probably when the Cracow physicists will have the opportunity to see if the artificial intelligence they have created will do its job in reconstructing particle tracks. Confirmation of its effectiveness in the conditions of a real experiment could mark the beginning of a new era in particle detection techniques.

Reference: Machine Learning based Event Reconstruction for the MUonE Experiment by Miosz Zdyba, Marcin Kucharczyk and Marcin Wolter, 10 March 2024, Computer Science. DOI: 10.7494/csci.2024.25.1.5690

The work of the team of physicists from the IFJ PAN was funded by a grant from the Polish National Science Centre.

See the original post:
Unveiling New Physics With AI-Powered Particle Tracking - SciTechDaily